Host Unit：Nanjing University of Information Science & Technology
Editor in chief：PAN Chengsheng
- Current Issue
- Online First
- Archive
- Most Downloaded
- Special Issue
- Special Column
Select AllDeselectExport15
Display Method:
HUANG Wenlong, ZHAO Haohao, KANG Jian, ZHI Xiaodong, WANG Dongchuan, ZHOU Weixun, NI Huan, GUAN Haiyan
2025,17(2):151-164, DOI: 10.13878/j.cnki.jnuist.20241009001
Abstract:
To accurately detect road damages with large size differences and small scales in vehicle-mounted images,this paper presents a real-time road damage detection model based on improved YOLOv5s,termed as VRD-YOLO (Vehicle-mounted image Road Damage Detection-YOLO).Firstly,a Channel Mix Slide Transformer (CMST) module is proposed to enhance the model's global context modeling capability and strengthen the extraction of fine-grained road damage semantic feature information.Secondly,a generalized feature pyramid with cross-layer fusion and cross-scale fusion is introduced to expand the network receptive field and strengthen the fusion of multi-scale damage features.Thirdly,to optimize the model's feature response and further improve detection performance,a dynamic detection head is designed to achieve scale perception,spatial perception,and task perception.Finally,a Vehicle-mounted Image Road Damage Dataset (VIRDD) is constructed to expand the quantity and types of existing road damage datasets,and ablation and comparative experiments are conducted based on this dataset.Experimental results show that the VRD-YOLO achieves a mean Average Precision (mAP@0.5) of 74.45% on the VIRDD dataset,with a detection speed reaching 28.56 frames per second.Compared to the YOLOv5s model,VRD-YOLO improves the precision,recall,F1 score,and mAP by 2.79,2.32,2.54,and 3.19 percentage points,respectively.Additionally,compared with six other classical and mainstream object detection models,the proposed VRD-YOLO attains the highest detection accuracy with the smallest model parameter count of 9.68 million,verifying its effectiveness and superiority.
To accurately detect road damages with large size differences and small scales in vehicle-mounted images,this paper presents a real-time road damage detection model based on improved YOLOv5s,termed as VRD-YOLO (Vehicle-mounted image Road Damage Detection-YOLO).Firstly,a Channel Mix Slide Transformer (CMST) module is proposed to enhance the model's global context modeling capability and strengthen the extraction of fine-grained road damage semantic feature information.Secondly,a generalized feature pyramid with cross-layer fusion and cross-scale fusion is introduced to expand the network receptive field and strengthen the fusion of multi-scale damage features.Thirdly,to optimize the model's feature response and further improve detection performance,a dynamic detection head is designed to achieve scale perception,spatial perception,and task perception.Finally,a Vehicle-mounted Image Road Damage Dataset (VIRDD) is constructed to expand the quantity and types of existing road damage datasets,and ablation and comparative experiments are conducted based on this dataset.Experimental results show that the VRD-YOLO achieves a mean Average Precision (mAP@0.5) of 74.45% on the VIRDD dataset,with a detection speed reaching 28.56 frames per second.Compared to the YOLOv5s model,VRD-YOLO improves the precision,recall,F1 score,and mAP by 2.79,2.32,2.54,and 3.19 percentage points,respectively.Additionally,compared with six other classical and mainstream object detection models,the proposed VRD-YOLO attains the highest detection accuracy with the smallest model parameter count of 9.68 million,verifying its effectiveness and superiority.
2025,17(2):165-171, DOI: 10.13878/j.cnki.jnuist.20241024001
Abstract:
To address the deficiencies of traditional graph neural network methods in point cloud semantic segmentation,such as high requirements for supervision accuracy,one-way only node label propagation,and neglect of global information,this paper proposes a point cloud semantic segmentation method based on bidirectional attention mechanism.Firstly,the point cloud is over-segmented into superpoints and a superpoint graph is constructed,thus introducing the point cloud classification problem into the superpoint graph network framework.Subsequently,the two-way attention module is utilized to alternately focus on superpoints and update their features according to the weights of neighboring superpoints,enabling the two-way information propagation.Unlike previous graph pooling methods,this study applies both maximum pooling and average pooling,and combines their pooled features.Finally,the public dataset Semantic 3D is used for training and experiments.The results show that the proposed method can effectively correct labelling errors while coupling local features with long-range information,and the mean Intersection over Union (mIoU) and overall Accuracy (oAcc) of the dataset are 75.4% and 95.1%,respectively,exhibiting a better label delivery mechanism and higher classification accuracy compared with existing methods.
To address the deficiencies of traditional graph neural network methods in point cloud semantic segmentation,such as high requirements for supervision accuracy,one-way only node label propagation,and neglect of global information,this paper proposes a point cloud semantic segmentation method based on bidirectional attention mechanism.Firstly,the point cloud is over-segmented into superpoints and a superpoint graph is constructed,thus introducing the point cloud classification problem into the superpoint graph network framework.Subsequently,the two-way attention module is utilized to alternately focus on superpoints and update their features according to the weights of neighboring superpoints,enabling the two-way information propagation.Unlike previous graph pooling methods,this study applies both maximum pooling and average pooling,and combines their pooled features.Finally,the public dataset Semantic 3D is used for training and experiments.The results show that the proposed method can effectively correct labelling errors while coupling local features with long-range information,and the mean Intersection over Union (mIoU) and overall Accuracy (oAcc) of the dataset are 75.4% and 95.1%,respectively,exhibiting a better label delivery mechanism and higher classification accuracy compared with existing methods.
2025,17(2):172-180, DOI: 10.13878/j.cnki.jnuist.20240927001
Abstract:
To tackle the current challenges of low efficiency,poor performance,and inadequate real-time capabilities in bridge crack detection,this paper introduces a drone-based image detection method for bridge cracks using an improved YOLOv8 model.Firstly,the dynamic snake convolution kernel is integrated into the C2f module in the backbone of YOLOv8 to enhance the crack feature extraction.Then,the Context Augmentation Module (CAM) is introduced to improve the detection capability for small targets.Finally,the influence of low-quality datasets on detection results is reduced via optimizing the prediction box loss function.Experimental results show that the improved model achieves a GFLOPs of 14.4 and a mean Average Precision (mAP@50) of 94%,exhibiting a significant accuracy improvement compared to the baseline models.The detection speed reaches 147 frames per second,satisfying the requirements for real-time crack detection by UAVs.
To tackle the current challenges of low efficiency,poor performance,and inadequate real-time capabilities in bridge crack detection,this paper introduces a drone-based image detection method for bridge cracks using an improved YOLOv8 model.Firstly,the dynamic snake convolution kernel is integrated into the C2f module in the backbone of YOLOv8 to enhance the crack feature extraction.Then,the Context Augmentation Module (CAM) is introduced to improve the detection capability for small targets.Finally,the influence of low-quality datasets on detection results is reduced via optimizing the prediction box loss function.Experimental results show that the improved model achieves a GFLOPs of 14.4 and a mean Average Precision (mAP@50) of 94%,exhibiting a significant accuracy improvement compared to the baseline models.The detection speed reaches 147 frames per second,satisfying the requirements for real-time crack detection by UAVs.
2025,17(2):181-190, DOI: 10.13878/j.cnki.jnuist.20240513002
Abstract:
Aiming at the extraction of cavern surface deformation from three-dimensional laser scanning dense point clouds,we propose a method integrating the Multiscale Model-to-Model Cloud Comparison (M3C2) with an improved Alpha Shapes algorithm.First,the two-phase surface point cloud data are registered,and the improved Alpha Shapes algorithm is used to identify the outer contour point clouds.After the fine registration of these two-phase outer contour point clouds,the M3C2 algorithm calculates the deformation value of each point,and finally the continuous deformation regions are extracted through distance clustering.Experimental results show that the proposed method effectively eliminates the points at small furrows as well as those affected by mixed pixels.Specifically,the removal rates of point clouds in the two phases within 10 m from the scanner to the cavern section are 14.17% and 13.52%,respectively,which are 6.25% and 6.42% within 70 m.This method accurately and efficiently extracts the cavern surface deformation regions with more than twice the registration error.
Aiming at the extraction of cavern surface deformation from three-dimensional laser scanning dense point clouds,we propose a method integrating the Multiscale Model-to-Model Cloud Comparison (M3C2) with an improved Alpha Shapes algorithm.First,the two-phase surface point cloud data are registered,and the improved Alpha Shapes algorithm is used to identify the outer contour point clouds.After the fine registration of these two-phase outer contour point clouds,the M3C2 algorithm calculates the deformation value of each point,and finally the continuous deformation regions are extracted through distance clustering.Experimental results show that the proposed method effectively eliminates the points at small furrows as well as those affected by mixed pixels.Specifically,the removal rates of point clouds in the two phases within 10 m from the scanner to the cavern section are 14.17% and 13.52%,respectively,which are 6.25% and 6.42% within 70 m.This method accurately and efficiently extracts the cavern surface deformation regions with more than twice the registration error.
2025,17(2):191-202, DOI: 10.13878/j.cnki.jnuist.20240229001
Abstract:
To address the low visual quality of stego-image in existing color image Reversible Data Hiding (RDH) algorithms,a novel RDH scheme utilizing multi-level interpolation prediction and global sorting is proposed.Firstly,to fully exploit the features of different texture regions in the image,a multi-level interpolation prediction method is designed to significantly improve the prediction accuracy of pixels.Then,a complexity-based global sorting strategy is designed to sort the prediction errors in the three channels of color images respectively,thereby fully utilizing the global characteristics of the prediction errors in each channel to generate a more concentrated Three-Dimensional Prediction Error Histogram (3D PEH).Finally,an adaptive 3D mapping strategy is used to modify the error histogram and embed secret data.Experimental results show that the proposed approach outperforms some of the latest schemes in embedding performance.
To address the low visual quality of stego-image in existing color image Reversible Data Hiding (RDH) algorithms,a novel RDH scheme utilizing multi-level interpolation prediction and global sorting is proposed.Firstly,to fully exploit the features of different texture regions in the image,a multi-level interpolation prediction method is designed to significantly improve the prediction accuracy of pixels.Then,a complexity-based global sorting strategy is designed to sort the prediction errors in the three channels of color images respectively,thereby fully utilizing the global characteristics of the prediction errors in each channel to generate a more concentrated Three-Dimensional Prediction Error Histogram (3D PEH).Finally,an adaptive 3D mapping strategy is used to modify the error histogram and embed secret data.Experimental results show that the proposed approach outperforms some of the latest schemes in embedding performance.
2025,17(2):203-214, DOI: 10.13878/j.cnki.jnuist.20230927001
Abstract:
Here,we propose a pruning and optimization approach based on Gradient Weight Pursuit (GWP) to address the overfitting in unsupervised domain,which manifests as significantly lower accuracy on downstream tasks compared to that on training sets.To tackle the overfitting challenge in unsupervised domain,we employ the dense-sparse-dense strategy,focusing on both difference-based and adversarial adaptive methods.First,the network is pretrained intensively to identify crucial connections.Second,during the pruning stage,the optimization algorithm in this paper distinguishes itself from original dense-sparse-dense strategy by jointly considering both weight and gradient information.Specifically,it leverages both weight (i.e.zero-order information) and gradient (i.e.first-order information) to influence pruning process.In the final dense phase,the pruned connections are restored and the dense network is retrained with a reduced learning rate.Finally,the obtained network achieves desirable outcomes in downstream tasks.The experimental results show that the proposed GWP approach can effectively improve the accuracy of downstream tasks,offering a plug-and-play capability compared with original difference-based and adversarial domain adaptation methods.
Here,we propose a pruning and optimization approach based on Gradient Weight Pursuit (GWP) to address the overfitting in unsupervised domain,which manifests as significantly lower accuracy on downstream tasks compared to that on training sets.To tackle the overfitting challenge in unsupervised domain,we employ the dense-sparse-dense strategy,focusing on both difference-based and adversarial adaptive methods.First,the network is pretrained intensively to identify crucial connections.Second,during the pruning stage,the optimization algorithm in this paper distinguishes itself from original dense-sparse-dense strategy by jointly considering both weight and gradient information.Specifically,it leverages both weight (i.e.zero-order information) and gradient (i.e.first-order information) to influence pruning process.In the final dense phase,the pruned connections are restored and the dense network is retrained with a reduced learning rate.Finally,the obtained network achieves desirable outcomes in downstream tasks.The experimental results show that the proposed GWP approach can effectively improve the accuracy of downstream tasks,offering a plug-and-play capability compared with original difference-based and adversarial domain adaptation methods.
2025,17(2):215-226, DOI: 10.13878/j.cnki.jnuist.20230810003
Abstract:
As a crucial component of weather systems,3D cloud simulation plays a significant role in various fields such as military and aviation.However,the current mainstream Bounding Volume Hierarchy (BVH) algorithm exhibits inefficient rendering performance when dealing with non-uniform and large-volume clouds.Here,a cloud rendering approach based on optimized BVH algorithm is proposed.The data points from the WRF(Weather Research and Forecasting) grids are used as cloud primitives,and a Z-order Hilbert curve is employed for spatial sorting.The BVH algorithm based on the Surface Area Heuristic (SAH) is optimized by locally optimizing the cloud primitive density,aiming to enhance computational efficiency.To tackle the data access overhead of overlapping BVH nodes,a novel storage structure called Overlapping Node Sets (ONS) is introduced,which reduces the time complexity.The optimized BVH algorithm reduces unnecessary intersection tests between rays and triangle surfaces,and resolves issues related to invalid boundary volume overlaps.Simulation experiments demonstrate that the proposed method achieves a 15.6％ improvement in SAH cost compared to similar state-of-the-art algorithms,a 10％ improvement in EPO(Effective Partial Overlap),and a reduction of over 100％ in construction time.The computational efficiency of the optimized BVH algorithm outperforms similar algorithms in any WRF cloud scenario,indicating its capability for rapid rendering of WRF cloud products.
As a crucial component of weather systems,3D cloud simulation plays a significant role in various fields such as military and aviation.However,the current mainstream Bounding Volume Hierarchy (BVH) algorithm exhibits inefficient rendering performance when dealing with non-uniform and large-volume clouds.Here,a cloud rendering approach based on optimized BVH algorithm is proposed.The data points from the WRF(Weather Research and Forecasting) grids are used as cloud primitives,and a Z-order Hilbert curve is employed for spatial sorting.The BVH algorithm based on the Surface Area Heuristic (SAH) is optimized by locally optimizing the cloud primitive density,aiming to enhance computational efficiency.To tackle the data access overhead of overlapping BVH nodes,a novel storage structure called Overlapping Node Sets (ONS) is introduced,which reduces the time complexity.The optimized BVH algorithm reduces unnecessary intersection tests between rays and triangle surfaces,and resolves issues related to invalid boundary volume overlaps.Simulation experiments demonstrate that the proposed method achieves a 15.6％ improvement in SAH cost compared to similar state-of-the-art algorithms,a 10％ improvement in EPO(Effective Partial Overlap),and a reduction of over 100％ in construction time.The computational efficiency of the optimized BVH algorithm outperforms similar algorithms in any WRF cloud scenario,indicating its capability for rapid rendering of WRF cloud products.
2025,17(2):227-234, DOI: 10.13878/j.cnki.jnuist.20230921003
Abstract:
Texture extraction,a pivotal task in computer vision,significantly influences the accuracy of texture classification.Traditional single-texture extraction methods often fail to accurately describe the characteristics of various textures.To address this issue,this paper proposes a texture extraction approach based on an Improved Position Local Binary Pattern (IPLBP) and Gabor filters.The proposed IPLBP enhances texture description capability by integrating texture position information into the LBP framework.Specifically,the IPLBP algorithm captures local texture nuances,while Gabor filters extract global texture attributes.Subsequently,these two complementary feature sets are fused and classified using Support Vector Machine (SVM).Experimental results demonstrate that the proposed approach exhibits excellent performance in texture material classification tasks.Notably,compared to traditional LBP algorithms,the IPLBP-Gabor filter approach more accurately discerns the subtle differences between diverse texture features,thereby enhancing texture classification accuracy.
Texture extraction,a pivotal task in computer vision,significantly influences the accuracy of texture classification.Traditional single-texture extraction methods often fail to accurately describe the characteristics of various textures.To address this issue,this paper proposes a texture extraction approach based on an Improved Position Local Binary Pattern (IPLBP) and Gabor filters.The proposed IPLBP enhances texture description capability by integrating texture position information into the LBP framework.Specifically,the IPLBP algorithm captures local texture nuances,while Gabor filters extract global texture attributes.Subsequently,these two complementary feature sets are fused and classified using Support Vector Machine (SVM).Experimental results demonstrate that the proposed approach exhibits excellent performance in texture material classification tasks.Notably,compared to traditional LBP algorithms,the IPLBP-Gabor filter approach more accurately discerns the subtle differences between diverse texture features,thereby enhancing texture classification accuracy.
2025,17(2):235-244, DOI: 10.13878/j.cnki.jnuist.20240617001
Abstract:
Here,a bearing fault diagnosis method based on recurrence analysis and Stacking ensemble learning is proposed to effectively extract nonlinear information from rolling bearing signals and improve diagnostic accuracy.Firstly,the nonlinear information in bearing signals is mapped to a two-dimensional recurrence plot through the application of recurrence analysis theory.Convolutional Neural Network (CNN) and Support Vector Machine (SVM) models are established from the perspectives of image recognition and recurrence quantification analysis,respectively.Finally,the Stacking method is employed to integrate these two models,leveraging their respective strengths.Experimental results demonstrate that the proposed method significantly improves the classification accuracy of bearing vibration signals and exhibits excellent stability under varying load conditions,providing a reliable solution for bearing fault diagnosis.
Here,a bearing fault diagnosis method based on recurrence analysis and Stacking ensemble learning is proposed to effectively extract nonlinear information from rolling bearing signals and improve diagnostic accuracy.Firstly,the nonlinear information in bearing signals is mapped to a two-dimensional recurrence plot through the application of recurrence analysis theory.Convolutional Neural Network (CNN) and Support Vector Machine (SVM) models are established from the perspectives of image recognition and recurrence quantification analysis,respectively.Finally,the Stacking method is employed to integrate these two models,leveraging their respective strengths.Experimental results demonstrate that the proposed method significantly improves the classification accuracy of bearing vibration signals and exhibits excellent stability under varying load conditions,providing a reliable solution for bearing fault diagnosis.
2025,17(2):245-255, DOI: 10.13878/j.cnki.jnuist.20240512002
Abstract:
In the field of Brain-Computer Interface (BCI),the recognition of natural hand movements through electroencephalography (EEG) is crucial for achieving natural and precise human-machine interaction.However,attempts to enhance model generalization ability across different subjects using transfer learning are still rare in studies focusing on natural hand movement paradigms.Here,we investigate three natural hand movement paradigms of grasping,pinching and twisting through EEG experiments,and validate the effectiveness of two transfer learning algorithms,namely CA-MDM(Covariance matrix centroid Alignment-Minimum Distance to Riemannian Mean) and CA-JDA(Covariance matrix centroid Alignment-Joint Distribution Adaptation),on our experimental dataset.The results show that CA-JDA achieves average accuracies of 60.51%±5.78% and 34.89%±4.42% in binary and quadruple classification tasks,respectively,while CA-MDM performs at 63.88%±4.59% and 35.71%±4.84% in the same tasks,highlighting the advantages of Riemannian space-based classifiers in handling covariance features.This study not only confirms the feasibility of transfer learning in natural hand movement paradigms but also aids in reducing calibration time for BCI systems and implementing natural human-machine interaction strategies.
In the field of Brain-Computer Interface (BCI),the recognition of natural hand movements through electroencephalography (EEG) is crucial for achieving natural and precise human-machine interaction.However,attempts to enhance model generalization ability across different subjects using transfer learning are still rare in studies focusing on natural hand movement paradigms.Here,we investigate three natural hand movement paradigms of grasping,pinching and twisting through EEG experiments,and validate the effectiveness of two transfer learning algorithms,namely CA-MDM(Covariance matrix centroid Alignment-Minimum Distance to Riemannian Mean) and CA-JDA(Covariance matrix centroid Alignment-Joint Distribution Adaptation),on our experimental dataset.The results show that CA-JDA achieves average accuracies of 60.51%±5.78% and 34.89%±4.42% in binary and quadruple classification tasks,respectively,while CA-MDM performs at 63.88%±4.59% and 35.71%±4.84% in the same tasks,highlighting the advantages of Riemannian space-based classifiers in handling covariance features.This study not only confirms the feasibility of transfer learning in natural hand movement paradigms but also aids in reducing calibration time for BCI systems and implementing natural human-machine interaction strategies.
2025,17(2):256-264, DOI: 10.13878/j.cnki.jnuist.20240506001
Abstract:
To address the issues of low search efficiency,long distance,and non-smooth paths in traditional path planning algorithms for autonomous vehicles,this study proposes an improvement by using key nodes of the optimized ant colony algorithm to replace the local target points in the dynamic window approach.Additionally,a target distance evaluation sub-function is incorporated into the dynamic window approach's evaluation function to enhance the efficiency and smoothness of path planning.Furthermore,a path decision-making method is employed to solve the problem of global path failure,enabling the vehicle to avoid obstacles and meet safety requirements of path planning.The improved ant colony algorithm utilizes the positional information of the start and end points to create an uneven initial pheromone distribution,thereby reducing time consumption during the initial search phase.By maintaining the global optimal paths and enhancing the pheromone concentration of excellent local paths,the pheromone update mechanism is optimized to speed up path exploration efficiency.The planned path is further optimized to reduce redundancy in nodes and turning points,thereby shortening path length.Simulation results show that compared to traditional path planning algorithms,the proposed integrated algorithm achieves better performance in terms of distance,smoothness,and convergence,aligning with the safety requirements for autonomous vehicle operation.
To address the issues of low search efficiency,long distance,and non-smooth paths in traditional path planning algorithms for autonomous vehicles,this study proposes an improvement by using key nodes of the optimized ant colony algorithm to replace the local target points in the dynamic window approach.Additionally,a target distance evaluation sub-function is incorporated into the dynamic window approach's evaluation function to enhance the efficiency and smoothness of path planning.Furthermore,a path decision-making method is employed to solve the problem of global path failure,enabling the vehicle to avoid obstacles and meet safety requirements of path planning.The improved ant colony algorithm utilizes the positional information of the start and end points to create an uneven initial pheromone distribution,thereby reducing time consumption during the initial search phase.By maintaining the global optimal paths and enhancing the pheromone concentration of excellent local paths,the pheromone update mechanism is optimized to speed up path exploration efficiency.The planned path is further optimized to reduce redundancy in nodes and turning points,thereby shortening path length.Simulation results show that compared to traditional path planning algorithms,the proposed integrated algorithm achieves better performance in terms of distance,smoothness,and convergence,aligning with the safety requirements for autonomous vehicle operation.
Select AllDeselectExport
Display Method:
Abstract:
Aiming at the problem of motion blur phenomenon in images taken from dynamic scenes, which leads to image quality degradation and serious loss of detail information, this paper proposes a multi-scale motion image deblurring method based on dual-domain feature fusion. First, a dual-domain feature fusion module is designed to extract spatial-domain features and frequency-domain features from blurred images in parallel using a two-branch structure, and the dual-domain features are deeply fused to improve the feature representation capability of the network model for high-frequency details. Then, a multi-scale feature aggregation module is designed to use cross-channel self-attention to aggregate the coded features of different scales of blurred images, and dynamically adjust the weights of different scales of feature maps to enhance the robustness of the model. Finally, the training loss function is improved, and the training of the network model is supervised using a joint multiscale loss function combining content loss, wavelet domain reconstruction loss and edge loss to further improve the deblurring effect. The results of comparison experiments and ablation experiments of this paper"s method on public datasets show that the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM metrics are 32.56 dB and 0.959, respectively, which are superior to other comparison methods. The experimental results show that this paper"s method can effectively improve the de-blurring effect and has good robustness.
Aiming at the problem of motion blur phenomenon in images taken from dynamic scenes, which leads to image quality degradation and serious loss of detail information, this paper proposes a multi-scale motion image deblurring method based on dual-domain feature fusion. First, a dual-domain feature fusion module is designed to extract spatial-domain features and frequency-domain features from blurred images in parallel using a two-branch structure, and the dual-domain features are deeply fused to improve the feature representation capability of the network model for high-frequency details. Then, a multi-scale feature aggregation module is designed to use cross-channel self-attention to aggregate the coded features of different scales of blurred images, and dynamically adjust the weights of different scales of feature maps to enhance the robustness of the model. Finally, the training loss function is improved, and the training of the network model is supervised using a joint multiscale loss function combining content loss, wavelet domain reconstruction loss and edge loss to further improve the deblurring effect. The results of comparison experiments and ablation experiments of this paper"s method on public datasets show that the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM metrics are 32.56 dB and 0.959, respectively, which are superior to other comparison methods. The experimental results show that this paper"s method can effectively improve the de-blurring effect and has good robustness.
Abstract:
Given the complex characteristics of carbon price series, such as nonlinearity and non-smoothness, the prediction is quite challenging. To improve the forecasting accuracy, TTBiGRUA was proposed as a carbon price prediction model, which combines time-varying filter empirical modal decomposition (TVFEMD), sample entropy (SE), bidirectional gated recurrent unit (BiGRU) and autoregressive integrated moving average (ARIMA) . First, the carbon price was decomposed into modal components with different frequencies by TVFEMD; then, the sample entropy was used to evaluate the complexity of each component and the K-means algorithm was used for reconstruction. Secondary TVFEMD decomposition was then applied to the most volatile modal components after reconstruction to further extract features and reduce modal aliasing. The high-frequency and low-frequency components were separated according to the sample entropy. The high-frequency components were predicted by BiGRU, while the low-frequency components were predicted by ARIMA, and the final prediction results were obtained by superimposing the component prediction results. We applied the actual carbon price data from the carbon markets in Guangdong and Hubei, China, and evaluated the predictive validity and robustness of the model by applying five evaluation indicators and the Diebold-Mariano (DM) test. The results show that the predictive accuracy of the proposed model is better than other benchmark comparison models.
Given the complex characteristics of carbon price series, such as nonlinearity and non-smoothness, the prediction is quite challenging. To improve the forecasting accuracy, TTBiGRUA was proposed as a carbon price prediction model, which combines time-varying filter empirical modal decomposition (TVFEMD), sample entropy (SE), bidirectional gated recurrent unit (BiGRU) and autoregressive integrated moving average (ARIMA) . First, the carbon price was decomposed into modal components with different frequencies by TVFEMD; then, the sample entropy was used to evaluate the complexity of each component and the K-means algorithm was used for reconstruction. Secondary TVFEMD decomposition was then applied to the most volatile modal components after reconstruction to further extract features and reduce modal aliasing. The high-frequency and low-frequency components were separated according to the sample entropy. The high-frequency components were predicted by BiGRU, while the low-frequency components were predicted by ARIMA, and the final prediction results were obtained by superimposing the component prediction results. We applied the actual carbon price data from the carbon markets in Guangdong and Hubei, China, and evaluated the predictive validity and robustness of the model by applying five evaluation indicators and the Diebold-Mariano (DM) test. The results show that the predictive accuracy of the proposed model is better than other benchmark comparison models.
Abstract:
In order to improve the efficiency of substation inspection, a detection and recognition method of distorted terminal logo text in substation based on deep learning was proposed. The terminal identification detection model uses DBNet as the basic framework, replaces the original ResNet backbone network with ConvNeXt V2, and uses its modern architecture design to significantly improve the global information modeling ability of the model and the feature extraction ability of the terminal identification. In order to further improve the detection accuracy of distorted terminal signs, an efficient multi-scale attention module EMA and a deformable convolutional network DCNv4 are introduced to effectively improve the ability to capture global context information and enhance the robustness to irregular text shapes. After optimization, the detection accuracy of the terminal sign detection model reaches 97.4%, which is 18.9 percentage points higher than the original model. And the amount of calculation is only increased by 2.9%. The terminal identification model takes CRNN as the basic framework, and introduces spatial and channel convolution SCConv to optimize the feature extraction process, which significantly reduces redundant features and reduces the computational burden. In the sequence modeling part, the selective state space network Mamba is used to replace the original LSTM, its state space model and selective mechanism can dynamically model the sequence data, adaptively focus on the important part of the sequence, and significantly enhance the ability to capture long sequence dependencies. The recognition accuracy of the optimized terminal identification model reaches 98.2%. It is 2.7 percentage points higher than the original model. Experimental results show that the proposed method has excellent detection ability and recognition accuracy for negative factors such as distortion, weak light and blur in substation terminal identification.
In order to improve the efficiency of substation inspection, a detection and recognition method of distorted terminal logo text in substation based on deep learning was proposed. The terminal identification detection model uses DBNet as the basic framework, replaces the original ResNet backbone network with ConvNeXt V2, and uses its modern architecture design to significantly improve the global information modeling ability of the model and the feature extraction ability of the terminal identification. In order to further improve the detection accuracy of distorted terminal signs, an efficient multi-scale attention module EMA and a deformable convolutional network DCNv4 are introduced to effectively improve the ability to capture global context information and enhance the robustness to irregular text shapes. After optimization, the detection accuracy of the terminal sign detection model reaches 97.4%, which is 18.9 percentage points higher than the original model. And the amount of calculation is only increased by 2.9%. The terminal identification model takes CRNN as the basic framework, and introduces spatial and channel convolution SCConv to optimize the feature extraction process, which significantly reduces redundant features and reduces the computational burden. In the sequence modeling part, the selective state space network Mamba is used to replace the original LSTM, its state space model and selective mechanism can dynamically model the sequence data, adaptively focus on the important part of the sequence, and significantly enhance the ability to capture long sequence dependencies. The recognition accuracy of the optimized terminal identification model reaches 98.2%. It is 2.7 percentage points higher than the original model. Experimental results show that the proposed method has excellent detection ability and recognition accuracy for negative factors such as distortion, weak light and blur in substation terminal identification.
Abstract:
Aiming at the path planning problem of multiple unmanned vehicles in the real environment, an algorithm design scheme is proposed under the framework of Multi-Agent Soft Actor-Critic (MASAC). To optimize and verify the performance of the algorithm, this paper optimizes the algorithm from three aspects. Firstly, based on the potential shaping reward technology, a dense reward function is designed to provide more abundant, timely and effective feedback signals for the learning process of the algorithm, thus significantly accelerating the convergence speed of the algorithm. Secondly, the traditional experience replay buffer is improved by using the double consecutive frame technology. This technology incorporates two consecutive frames of observation data as a whole unit into the experience replay buffer, effectively capturing the dynamic information of environmental state changes and improving the training efficiency and stability. Thirdly, relying on the Gazebo simulation platform, a highly realistic dynamic obstacle environment is built, which provides a rich variety of and extremely challenging training samples for the training of the algorithm, ensuring that the algorithm can be fully learned and optimized under simulated real conditions. Finally, the effectiveness of the algorithm is verified through ablation experiments and robustness tests.
Aiming at the path planning problem of multiple unmanned vehicles in the real environment, an algorithm design scheme is proposed under the framework of Multi-Agent Soft Actor-Critic (MASAC). To optimize and verify the performance of the algorithm, this paper optimizes the algorithm from three aspects. Firstly, based on the potential shaping reward technology, a dense reward function is designed to provide more abundant, timely and effective feedback signals for the learning process of the algorithm, thus significantly accelerating the convergence speed of the algorithm. Secondly, the traditional experience replay buffer is improved by using the double consecutive frame technology. This technology incorporates two consecutive frames of observation data as a whole unit into the experience replay buffer, effectively capturing the dynamic information of environmental state changes and improving the training efficiency and stability. Thirdly, relying on the Gazebo simulation platform, a highly realistic dynamic obstacle environment is built, which provides a rich variety of and extremely challenging training samples for the training of the algorithm, ensuring that the algorithm can be fully learned and optimized under simulated real conditions. Finally, the effectiveness of the algorithm is verified through ablation experiments and robustness tests.
Abstract:
A vialess polarization rotation transmission metasurface (PRTM) with high frequency selectivity is proposed, which can rotate the polarization direction of the linearly polarized incident wave by 90° and has excellent frequency selection characteristics and angular stability. The PRTM unit cell has been implemented with three metallic layers which are separated by two substrate layers, in which the top and bottom metallic layers are the polarizing grids which are perpendicular to each other. Meanwhile, the middle metallic layer is consisted of the array of dual-mode rectangular ring which is miniaturized with four patches additionally loaded at the corners. The PRTM unit cell has two cross-polarized transmission poles and one cross-polarized transmission null, where the cross-polarized transmission zero position is flexibly adjustable.Due to the introduction of the cross-polarized transmission null, the upper or lower stop band of the PRTM unit cell can be achieved with excellent frequency selection characteristics. Then the physical size parameters of the PRTM unit cell has been studied, and a set of design processes of the PRTM unit cell are summarized. Finally, two PRTM prototypes with the center frequency of 15GHz have been fabricated and measured, and measurement results were in good agreement with the simulation results.
A vialess polarization rotation transmission metasurface (PRTM) with high frequency selectivity is proposed, which can rotate the polarization direction of the linearly polarized incident wave by 90° and has excellent frequency selection characteristics and angular stability. The PRTM unit cell has been implemented with three metallic layers which are separated by two substrate layers, in which the top and bottom metallic layers are the polarizing grids which are perpendicular to each other. Meanwhile, the middle metallic layer is consisted of the array of dual-mode rectangular ring which is miniaturized with four patches additionally loaded at the corners. The PRTM unit cell has two cross-polarized transmission poles and one cross-polarized transmission null, where the cross-polarized transmission zero position is flexibly adjustable.Due to the introduction of the cross-polarized transmission null, the upper or lower stop band of the PRTM unit cell can be achieved with excellent frequency selection characteristics. Then the physical size parameters of the PRTM unit cell has been studied, and a set of design processes of the PRTM unit cell are summarized. Finally, two PRTM prototypes with the center frequency of 15GHz have been fabricated and measured, and measurement results were in good agreement with the simulation results.
Abstract:
Aiming at the problem of ineffective emotion recognition in virtual sports, this study proposes a multimodal fusion emotion recognition algorithm based on facial images and heart rate variability (MFER-FIHRV). By capturing and analyzing data from both facial image and heart rate variability (HRV) modalities, the proposed approach enables a fine-grained perception of users’ emotional changes, thereby facilitating a personalized human-computer interaction (HCI) experience tailored to their current emotional state. The algorithm first employs a multimodal fusion Transformer to achieve complementary learning between facial images and HRV data. Then, a multimodal feature fusion strategy is utilized to concatenate fused features with the original feature representations. Additionally, a lightweight self-attention mechanism is introduced to capture high-level semantic representations within the multimodal data. Extensive experiments were conducted on two public datasets, demonstrating the superior performance of the proposed method. The results confirm its effectiveness and robustness, providing valuable insights for the design and development of user experience systems.
Aiming at the problem of ineffective emotion recognition in virtual sports, this study proposes a multimodal fusion emotion recognition algorithm based on facial images and heart rate variability (MFER-FIHRV). By capturing and analyzing data from both facial image and heart rate variability (HRV) modalities, the proposed approach enables a fine-grained perception of users’ emotional changes, thereby facilitating a personalized human-computer interaction (HCI) experience tailored to their current emotional state. The algorithm first employs a multimodal fusion Transformer to achieve complementary learning between facial images and HRV data. Then, a multimodal feature fusion strategy is utilized to concatenate fused features with the original feature representations. Additionally, a lightweight self-attention mechanism is introduced to capture high-level semantic representations within the multimodal data. Extensive experiments were conducted on two public datasets, demonstrating the superior performance of the proposed method. The results confirm its effectiveness and robustness, providing valuable insights for the design and development of user experience systems.
Abstract:
Regarding the active lane-changing scenarios of intelligent vehicles in multi-lane urban road traffic situations，the sole use of a single control algorithm for active lane-changing path tracking may lead to issues such as reduced path tracking accuracy or vehicle handling stability. This paper proposes a decoupled lateral and longitudinal lane-changing trajectory tracking control strategy. Firstly, the trajectory tracking control is decoupled into lateral position tracking and longitudinal speed control. Subsequently, a feedforward plus feedback lateral controller is designed based on the discrete LQR principle, and a position and speed longitudinal controller is designed based on the dual PID principle, ultimately achieving tracking control of the planned trajectory. Finally, a co-simulation model is established using Matlab/Simulink, PreScan, and CarSim platforms for simulation testing. The results show that the decoupled lateral and longitudinal trajectory tracking control algorithm performs well in tracking the trajectory under different lane-changing conditions of the vehicle, with the tracking control errors of the vehicle all within the specified constraints. Moreover, compared to the MPC controller, it has a smaller centroid sideslip angle and yaw rate, offering better driving stability and comfort.
Regarding the active lane-changing scenarios of intelligent vehicles in multi-lane urban road traffic situations，the sole use of a single control algorithm for active lane-changing path tracking may lead to issues such as reduced path tracking accuracy or vehicle handling stability. This paper proposes a decoupled lateral and longitudinal lane-changing trajectory tracking control strategy. Firstly, the trajectory tracking control is decoupled into lateral position tracking and longitudinal speed control. Subsequently, a feedforward plus feedback lateral controller is designed based on the discrete LQR principle, and a position and speed longitudinal controller is designed based on the dual PID principle, ultimately achieving tracking control of the planned trajectory. Finally, a co-simulation model is established using Matlab/Simulink, PreScan, and CarSim platforms for simulation testing. The results show that the decoupled lateral and longitudinal trajectory tracking control algorithm performs well in tracking the trajectory under different lane-changing conditions of the vehicle, with the tracking control errors of the vehicle all within the specified constraints. Moreover, compared to the MPC controller, it has a smaller centroid sideslip angle and yaw rate, offering better driving stability and comfort.
Abstract:
Bayesian Network and SBAS-InSAR have been widely used in landslide susceptibility evaluation. However, how to give full play to the advantages of the two methods and integrate the two methods to further improve landslide susceptibility evaluation remains to be explored. Therefore, this paper takes Hanyuan County, where landslides frequently occur, as the study area. According to topographic and meteorological data, 20 traditional landslide factors are extracted, and according to SAR satellite data, the surface deformation rate is extracted using SBAS-InSAR technology as a deformation factor to participate in the calculation. After using the GeoDetector to screen the dominant factors, the Bayesian Network structure is constructed using the hill climbing algorithm combined with expert knowledge in the case of deformation factors, removal of deformation factors, and supplementation of one factor. The landslide susceptibility calculation is performed after parameter learning using the maximum likelihood estimation algorithm. The results show that the AUC value of the Bayesian Network model with deformation factors added reaches 0.957, which is 8.63% and 8.26% higher than that of the Bayesian Network without deformation factors and supplementation of one factor, respectively, and the susceptibility zoning results are more reasonable. The landslide susceptibility evaluation method provided in this paper can provide guidance for regional disaster prevention and mitigation, and provide new ideas for related research.
Bayesian Network and SBAS-InSAR have been widely used in landslide susceptibility evaluation. However, how to give full play to the advantages of the two methods and integrate the two methods to further improve landslide susceptibility evaluation remains to be explored. Therefore, this paper takes Hanyuan County, where landslides frequently occur, as the study area. According to topographic and meteorological data, 20 traditional landslide factors are extracted, and according to SAR satellite data, the surface deformation rate is extracted using SBAS-InSAR technology as a deformation factor to participate in the calculation. After using the GeoDetector to screen the dominant factors, the Bayesian Network structure is constructed using the hill climbing algorithm combined with expert knowledge in the case of deformation factors, removal of deformation factors, and supplementation of one factor. The landslide susceptibility calculation is performed after parameter learning using the maximum likelihood estimation algorithm. The results show that the AUC value of the Bayesian Network model with deformation factors added reaches 0.957, which is 8.63% and 8.26% higher than that of the Bayesian Network without deformation factors and supplementation of one factor, respectively, and the susceptibility zoning results are more reasonable. The landslide susceptibility evaluation method provided in this paper can provide guidance for regional disaster prevention and mitigation, and provide new ideas for related research.
Multi-branch mapping convolutional neural network for Alzheimer's disease image classification study
Abstract:
To address the classification problem of AD and cognitively normal (CN) patients, this study proposes a multi-atlas feature fusion model. First, the input brain imaging data undergo standardization and are then parcellated using different brain atlas templates. Fiber tractography is performed to compute adjacency matrices, constructing graph-structured data that incorporate topological information. To fully capture brain network characteristics, this study introduces a graph convolution module capable of extracting high-order topological features, along with a graph pooling module that reduces computational cost while preserving essential structural information. Finally, a multi-branch feature fusion framework is employed to effectively integrate features from different brain atlases, enhancing the model's generalization ability. Experimental results demonstrate that the proposed model achieves effective feature fusion across multiple brain atlas templates, efficiently extracting both local and global brain network features to improve classification accuracy. Additionally, the proposed graph convolution and graph pooling modules not only enhance classification performance but also significantly reduce computational costs, offering new insights into graph-based neuroimaging analysis.
To address the classification problem of AD and cognitively normal (CN) patients, this study proposes a multi-atlas feature fusion model. First, the input brain imaging data undergo standardization and are then parcellated using different brain atlas templates. Fiber tractography is performed to compute adjacency matrices, constructing graph-structured data that incorporate topological information. To fully capture brain network characteristics, this study introduces a graph convolution module capable of extracting high-order topological features, along with a graph pooling module that reduces computational cost while preserving essential structural information. Finally, a multi-branch feature fusion framework is employed to effectively integrate features from different brain atlases, enhancing the model's generalization ability. Experimental results demonstrate that the proposed model achieves effective feature fusion across multiple brain atlas templates, efficiently extracting both local and global brain network features to improve classification accuracy. Additionally, the proposed graph convolution and graph pooling modules not only enhance classification performance but also significantly reduce computational costs, offering new insights into graph-based neuroimaging analysis.
Abstract:
Existing multimodal Visual Question Answering (VQA) models ignore the fine-grained interaction between local salient information in images and local basic words in texts, and the semantic relevance between images and texts needs to be improved. To this end, this paper proposes a multimodal visual question answering method based on fine-grained feature enhancement. First, a fine-grained feature extraction method is added to vision and text respectively to extract the semantic features of images and questions more comprehensively and accurately; then, in order to utilize the alignment information between modalities at different levels, an alignment-guided self-attention module is proposed to align the correspondence between fine-grained features and global semantic features within a single modality (visual or text), and fuse unimodal information at different levels in a unified way; finally, experiments are conducted on VQA v2.0 and VQA-CP v2 datasets. The results show that the proposed method performs better than existing models in various visual question answering evaluation indicators.
Existing multimodal Visual Question Answering (VQA) models ignore the fine-grained interaction between local salient information in images and local basic words in texts, and the semantic relevance between images and texts needs to be improved. To this end, this paper proposes a multimodal visual question answering method based on fine-grained feature enhancement. First, a fine-grained feature extraction method is added to vision and text respectively to extract the semantic features of images and questions more comprehensively and accurately; then, in order to utilize the alignment information between modalities at different levels, an alignment-guided self-attention module is proposed to align the correspondence between fine-grained features and global semantic features within a single modality (visual or text), and fuse unimodal information at different levels in a unified way; finally, experiments are conducted on VQA v2.0 and VQA-CP v2 datasets. The results show that the proposed method performs better than existing models in various visual question answering evaluation indicators.
Abstract:
Parameter identification plays a crucial role in the monitoring and management of transmission systems. However, current technology faces the following limitations: (1) Traditional methods only focus on the data of a single branch, ignoring the information of adjacent branches in the overall power grid topology; (2) The data pollution caused by external factors, such as data loss and noise, has a negative impact on the accuracy of parameter identification. In response to the above challenges, this article constructs a graph neural network model with GraphSAGE as the main body. This model integrates noise filtering modules and multi task loss training strategies, fully utilizing the advantages of graph convolutional networks to achieve efficient identification of transmission system parameters. This model can not only capture the topology information of the power grid, achieve joint identification of multiple branches, but also effectively handle data loss and noise, improving the robustness of the model. The experimental results show that compared to traditional methods, our GraphSAGE model has achieved significant improvements in the accuracy and stability of line parameter identification.
Parameter identification plays a crucial role in the monitoring and management of transmission systems. However, current technology faces the following limitations: (1) Traditional methods only focus on the data of a single branch, ignoring the information of adjacent branches in the overall power grid topology; (2) The data pollution caused by external factors, such as data loss and noise, has a negative impact on the accuracy of parameter identification. In response to the above challenges, this article constructs a graph neural network model with GraphSAGE as the main body. This model integrates noise filtering modules and multi task loss training strategies, fully utilizing the advantages of graph convolutional networks to achieve efficient identification of transmission system parameters. This model can not only capture the topology information of the power grid, achieve joint identification of multiple branches, but also effectively handle data loss and noise, improving the robustness of the model. The experimental results show that compared to traditional methods, our GraphSAGE model has achieved significant improvements in the accuracy and stability of line parameter identification.
Abstract:
Reconfigurable Intelligent Surfaces (RIS) can intelligently manipulate wireless channels to provide auxiliary transmission paths for communication systems. However, due to the "multiplicative fading" effect inherent in passive RIS (PRIS), existing PRIS-assisted hybrid cooperative systems can only achieve limited capacity gains in scenarios with weak direct links. To overcome this limitation, this paper introduces active RIS (ARIS) technology, leveraging its advantages such as flexible active beamforming and signal control, relatively low overall power consumption, and cost-effectiveness. Considering the stable reliability and widespread deployment of traditional relays, we investigate a hybrid system integrating ARIS with decode-and-forward (DF) relay in a multiple-input multiple-output (MIMO) framework. To maximize the system’s achievable rate, we formulate a joint optimization problem involving the source transmit beamforming matrix, relay transmit beamforming matrix, and ARIS active beamforming matrix. A joint optimization algorithm based on alternating optimization (AO) and fractional programming (FP) is proposed, decoupling the original problem into multiple standard quadratically constrained quadratic programming (QCQP) subproblems, which are then solved using the Lagrangian multiplier method. Simulation results demonstrate that the proposed system significantly outperforms benchmark schemes, with performance advantages becoming more pronounced as the number of RIS reflective elements increases.
Reconfigurable Intelligent Surfaces (RIS) can intelligently manipulate wireless channels to provide auxiliary transmission paths for communication systems. However, due to the "multiplicative fading" effect inherent in passive RIS (PRIS), existing PRIS-assisted hybrid cooperative systems can only achieve limited capacity gains in scenarios with weak direct links. To overcome this limitation, this paper introduces active RIS (ARIS) technology, leveraging its advantages such as flexible active beamforming and signal control, relatively low overall power consumption, and cost-effectiveness. Considering the stable reliability and widespread deployment of traditional relays, we investigate a hybrid system integrating ARIS with decode-and-forward (DF) relay in a multiple-input multiple-output (MIMO) framework. To maximize the system’s achievable rate, we formulate a joint optimization problem involving the source transmit beamforming matrix, relay transmit beamforming matrix, and ARIS active beamforming matrix. A joint optimization algorithm based on alternating optimization (AO) and fractional programming (FP) is proposed, decoupling the original problem into multiple standard quadratically constrained quadratic programming (QCQP) subproblems, which are then solved using the Lagrangian multiplier method. Simulation results demonstrate that the proposed system significantly outperforms benchmark schemes, with performance advantages becoming more pronounced as the number of RIS reflective elements increases.
Abstract:
To address the challenges of heavy guidewire manipulation tasks and insufficient automation in existing robotic systems for endovascular surgeries, this paper proposes an autonomous guidewire control algo-rithm based on diffusion policy, which enhances the success rate and stability of surgical robots in per-forming such underactuated control tasks. We first models the guidewire control task as a Markov Decision Process (MDP), then employs imitation learning to train the robotic system using expert demonstration data, enabling efficient operation across various vascular environments. The expert control strategy is sub-sequently modeled as a conditional probability distribution through a denoising diffusion probabilistic model, which guides control action generation based on observed states. Simulation tests in an aortic arch environment were conducted for both left subclavian artery and brachiocephalic trunk artery branch inter-ventions, with comparative analysis against existing imitation learning methods. Experimental results demonstrate that the proposed method achieves significant performance improvements in both tasks, ex-hibits low sensitivity to hyperparameter settings, and maintains superior training stability.
To address the challenges of heavy guidewire manipulation tasks and insufficient automation in existing robotic systems for endovascular surgeries, this paper proposes an autonomous guidewire control algo-rithm based on diffusion policy, which enhances the success rate and stability of surgical robots in per-forming such underactuated control tasks. We first models the guidewire control task as a Markov Decision Process (MDP), then employs imitation learning to train the robotic system using expert demonstration data, enabling efficient operation across various vascular environments. The expert control strategy is sub-sequently modeled as a conditional probability distribution through a denoising diffusion probabilistic model, which guides control action generation based on observed states. Simulation tests in an aortic arch environment were conducted for both left subclavian artery and brachiocephalic trunk artery branch inter-ventions, with comparative analysis against existing imitation learning methods. Experimental results demonstrate that the proposed method achieves significant performance improvements in both tasks, ex-hibits low sensitivity to hyperparameter settings, and maintains superior training stability.
Abstract:
Accurate and reliable wind speed prediction is the key to efficient utilization of wind energy. In consideration of the inherent instability of wind speed, a SWD-Transformer-KAN prediction model is proposed by fusing Swarm Decomposition (SWD), Transformer, and Kolmogorov-Arnold Network (KAN). Firstly, the original wind speed data is decomposed using the SWD method to extract key features. Secondly, a Transformer-KAN prediction model is established for each decomposed subsequence, which fully utilizes the temporal processing capability of Transformer and the nonlinear approximation capability of KAN. Finally, the resulting wind speed prediction values are obtained by superimposing the predicted results of all subsequences. To verify the effectiveness of the proposed model, it is experimentally compared with other models. The experimental results show that the SWD-Transformer-KAN model has the best predictive performance, with a determination coefficient as high as 99.91%.
Accurate and reliable wind speed prediction is the key to efficient utilization of wind energy. In consideration of the inherent instability of wind speed, a SWD-Transformer-KAN prediction model is proposed by fusing Swarm Decomposition (SWD), Transformer, and Kolmogorov-Arnold Network (KAN). Firstly, the original wind speed data is decomposed using the SWD method to extract key features. Secondly, a Transformer-KAN prediction model is established for each decomposed subsequence, which fully utilizes the temporal processing capability of Transformer and the nonlinear approximation capability of KAN. Finally, the resulting wind speed prediction values are obtained by superimposing the predicted results of all subsequences. To verify the effectiveness of the proposed model, it is experimentally compared with other models. The experimental results show that the SWD-Transformer-KAN model has the best predictive performance, with a determination coefficient as high as 99.91%.
Abstract:
Underdetermined blind source separation (UBSS) is a difficult problem in the field of blind signal processing, and the mixed matrix estimation is a key step in determining the success or failure of over-complete independent component analysis (ICA) and underdetermined blind separation. In order to improve the estimation accuracy of the underdetermined mixed matrix, a linear clustering analysis method for optimization of artificial bee colony search strategy is proposed based on single-source-point detection. Firstly, the observed signals in the time domain are transformed into the time-frequency domain using short-time Fourier transform (STFT), and single-source-point (SSP) detection is performed to enhance the linear clustering characteristics of the signal. Then, based on sparse time-frequency signals, the bee colony food source is matrix coded to make the artificial bee colony algorithm fit seamlessly with the blind separation problem. Randomness and deterministic search strategies are combined to coordinate the diversity of bee colonies with the convergence rate of clustering algorithm. Levy flight strategy is introduced into the local search of bee colonies to further explore the neighborhood of the current optimal solution to improve the clustering accuracy. Finally, the column vectors of the mixed matrix are estimated using the straight line direction vectors of the linear clustering generated by the artificial bee colony algorithm. The simulation results of audio signals show that the improved artificial bee colony search strategy proposed in this paper can not only provide effective linear clustering analysis, but also improve the estimation accuracy of underdetermined mixing matrix.
Underdetermined blind source separation (UBSS) is a difficult problem in the field of blind signal processing, and the mixed matrix estimation is a key step in determining the success or failure of over-complete independent component analysis (ICA) and underdetermined blind separation. In order to improve the estimation accuracy of the underdetermined mixed matrix, a linear clustering analysis method for optimization of artificial bee colony search strategy is proposed based on single-source-point detection. Firstly, the observed signals in the time domain are transformed into the time-frequency domain using short-time Fourier transform (STFT), and single-source-point (SSP) detection is performed to enhance the linear clustering characteristics of the signal. Then, based on sparse time-frequency signals, the bee colony food source is matrix coded to make the artificial bee colony algorithm fit seamlessly with the blind separation problem. Randomness and deterministic search strategies are combined to coordinate the diversity of bee colonies with the convergence rate of clustering algorithm. Levy flight strategy is introduced into the local search of bee colonies to further explore the neighborhood of the current optimal solution to improve the clustering accuracy. Finally, the column vectors of the mixed matrix are estimated using the straight line direction vectors of the linear clustering generated by the artificial bee colony algorithm. The simulation results of audio signals show that the improved artificial bee colony search strategy proposed in this paper can not only provide effective linear clustering analysis, but also improve the estimation accuracy of underdetermined mixing matrix.
Abstract:
A novel text image generation method based on diffusion model is proposed to address the problems of low image fidelity, difficult image generation operations, and limited applicability to specific task scenarios in existing text image generation methods. This method takes the current mainstream diffusion model as the main network, designs a new structure of residual blocks, effectively improves the model generation performance, adds a CBAM (Convolutional Block Attention Module) attention module to improve the noise estimation network, enhances the model"s ability to extract key information from images, further improves the quality of generated images, and finally combines conditional control networks to effectively achieve text image generation for specific poses. Qualitative and quantitative analysis, as well as ablation experiments, were conducted on the dataset using leading methods KNN Difsuion, CogView2, textStyleGAN, and simplefifusion using CelebA HQ. According to the evaluation indicators and generation results, the method can effectively improve the quality of text generated images, with an average decrease of 36.4% in FID, an average increase of 11.4% and 3.9% in IS and structural similarity. Combined with a conditional control network, the task of generating text images with directed actions is achieved.
A novel text image generation method based on diffusion model is proposed to address the problems of low image fidelity, difficult image generation operations, and limited applicability to specific task scenarios in existing text image generation methods. This method takes the current mainstream diffusion model as the main network, designs a new structure of residual blocks, effectively improves the model generation performance, adds a CBAM (Convolutional Block Attention Module) attention module to improve the noise estimation network, enhances the model"s ability to extract key information from images, further improves the quality of generated images, and finally combines conditional control networks to effectively achieve text image generation for specific poses. Qualitative and quantitative analysis, as well as ablation experiments, were conducted on the dataset using leading methods KNN Difsuion, CogView2, textStyleGAN, and simplefifusion using CelebA HQ. According to the evaluation indicators and generation results, the method can effectively improve the quality of text generated images, with an average decrease of 36.4% in FID, an average increase of 11.4% and 3.9% in IS and structural similarity. Combined with a conditional control network, the task of generating text images with directed actions is achieved.
Abstract:
Due to the diverse types and complex structures of wetlands, the classification and identification of wetlands by remote sensing is extremely challenging. In order to quickly and accurately identify wetland types by remote sensing, this paper takes the flood storage area of the Wanjiang River as the study area, constructs a sample set based on the spectral features, vegetation and water index features and texture features obtained from Sentinel-2 images, and introduces the deep learning compression-excitation and residual network (SE_ResNet) model to carry out the research on remote sensing identification of wetland types. The results show that the wetland type recognition of the SE_ResNet model is significantly better than the traditional supervised classification maximum likelihood method and the machine learning random forest method, and the overall accuracy is increased by 19.00 percentage points and 10.25 percentage points, respectively, to 94%, and the Kappa coefficient reaches 0.90. The SE_ResNet model can finely identify different wetland types, especially river wetlands, flood plain wetlands and freshwater lake wetlands, and the identification results are more refined and accurate than the global 30m wetland data product (GWL_FCS30) and the lake-type watershed natural-human integrated dataset (CODCLAB).
Due to the diverse types and complex structures of wetlands, the classification and identification of wetlands by remote sensing is extremely challenging. In order to quickly and accurately identify wetland types by remote sensing, this paper takes the flood storage area of the Wanjiang River as the study area, constructs a sample set based on the spectral features, vegetation and water index features and texture features obtained from Sentinel-2 images, and introduces the deep learning compression-excitation and residual network (SE_ResNet) model to carry out the research on remote sensing identification of wetland types. The results show that the wetland type recognition of the SE_ResNet model is significantly better than the traditional supervised classification maximum likelihood method and the machine learning random forest method, and the overall accuracy is increased by 19.00 percentage points and 10.25 percentage points, respectively, to 94%, and the Kappa coefficient reaches 0.90. The SE_ResNet model can finely identify different wetland types, especially river wetlands, flood plain wetlands and freshwater lake wetlands, and the identification results are more refined and accurate than the global 30m wetland data product (GWL_FCS30) and the lake-type watershed natural-human integrated dataset (CODCLAB).
Abstract:
Significant Wave Height (SWH) has complex nonlinear dynamic properties, which makes its accurate prediction a major challenge. Time-frequency decomposition technology is an effective means to deal with complex nonlinearities. However, existing methods do not consider the different time-frequency characteristics of the decomposed components of SWH. Therefore, this paper uses permutation entropy to divide the SWH components obtained after integrated empirical mode decomposition (EEMD) into high and low frequency categories, and constructs an optimized long short-term memory network - temporal convolutional network LSTM-TCN) based on their respective characteristics. Dual-channel temporal feature extraction module. Considering that different component prediction values have different impacts on the final SWH prediction result, the Bayesian average method (BMA) is introduced for weight allocation. Finally, this paper proposes a SWH dual-channel hybrid prediction model based on optimized deep learning. Experimental results show that compared with the existing state-of-the-art models, the evaluation indicators RMSE, MAE and MAPE of this model are significantly reduced in 1, 3, 6, and 12-hour SWH predictions, and it has better accuracy and stability.
Significant Wave Height (SWH) has complex nonlinear dynamic properties, which makes its accurate prediction a major challenge. Time-frequency decomposition technology is an effective means to deal with complex nonlinearities. However, existing methods do not consider the different time-frequency characteristics of the decomposed components of SWH. Therefore, this paper uses permutation entropy to divide the SWH components obtained after integrated empirical mode decomposition (EEMD) into high and low frequency categories, and constructs an optimized long short-term memory network - temporal convolutional network LSTM-TCN) based on their respective characteristics. Dual-channel temporal feature extraction module. Considering that different component prediction values have different impacts on the final SWH prediction result, the Bayesian average method (BMA) is introduced for weight allocation. Finally, this paper proposes a SWH dual-channel hybrid prediction model based on optimized deep learning. Experimental results show that compared with the existing state-of-the-art models, the evaluation indicators RMSE, MAE and MAPE of this model are significantly reduced in 1, 3, 6, and 12-hour SWH predictions, and it has better accuracy and stability.
Abstract:
The wrist is one of the most flexible parts of the human body. By decomposing the surface electromyography (sEMG) signals, the deep neural drive of human body movements can be effectively reversed. Currently, research on wrist moments is basically focused on isometric contractions, and substantial research is still needed for the decoding of neural drives under dynamic conditions. This paper studied the decomposition of motor units (MUs) when the wrist moves under different resistances. Specifically, different resistance levels were set by using a magnetorheological damper. The collected complete electromyography signals were divided into short intervals where the changes in motor unit action potentials (MUAPs) are minimal. Then, a classic decomposition algorithm was used in each short interval to obtain the motor unit spike train (MUST), and the MUs were tracked through the overlapping parts of the short intervals, so as to obtain the complete firing sequence. This paper studied the decomposition of motor units when the wrist extended and flexed under three resistances of 20% maximum voluntary contraction (MVC), 40% MVC, and 60% MVC. The results showed that under the three types of resistances, the dynamic decomposition algorithm proposed in this paper could effectively decompose motor units (MUs) from the forearm electromyographic signals, and as the resistance increased, the decomposition effect decreased to some extent. During wrist extension, up to 10 ± 1 MUs could be decomposed, and the pulse-to-noise ratio (PNR) and silhouette distance (SIL) could reach 19.87 ± 1.42 dB and 0.91 ± 0.03 respectively. During wrist flexion, up to 22 ± 3 MUs could be decomposed, and the PNR and SIL values could reach 20.69 ± 2.14 dB and 0.92 ± 0.03 respectively. This study indicates that it is feasible to perform neural decoding from the EMG of wrist movements under different resistances, which has important implications for the application of high-density surface EMG (HD-sEMG) under dynamic contractions.
The wrist is one of the most flexible parts of the human body. By decomposing the surface electromyography (sEMG) signals, the deep neural drive of human body movements can be effectively reversed. Currently, research on wrist moments is basically focused on isometric contractions, and substantial research is still needed for the decoding of neural drives under dynamic conditions. This paper studied the decomposition of motor units (MUs) when the wrist moves under different resistances. Specifically, different resistance levels were set by using a magnetorheological damper. The collected complete electromyography signals were divided into short intervals where the changes in motor unit action potentials (MUAPs) are minimal. Then, a classic decomposition algorithm was used in each short interval to obtain the motor unit spike train (MUST), and the MUs were tracked through the overlapping parts of the short intervals, so as to obtain the complete firing sequence. This paper studied the decomposition of motor units when the wrist extended and flexed under three resistances of 20% maximum voluntary contraction (MVC), 40% MVC, and 60% MVC. The results showed that under the three types of resistances, the dynamic decomposition algorithm proposed in this paper could effectively decompose motor units (MUs) from the forearm electromyographic signals, and as the resistance increased, the decomposition effect decreased to some extent. During wrist extension, up to 10 ± 1 MUs could be decomposed, and the pulse-to-noise ratio (PNR) and silhouette distance (SIL) could reach 19.87 ± 1.42 dB and 0.91 ± 0.03 respectively. During wrist flexion, up to 22 ± 3 MUs could be decomposed, and the PNR and SIL values could reach 20.69 ± 2.14 dB and 0.92 ± 0.03 respectively. This study indicates that it is feasible to perform neural decoding from the EMG of wrist movements under different resistances, which has important implications for the application of high-density surface EMG (HD-sEMG) under dynamic contractions.
Abstract:
Electric-hydrogen energy systems, which enable the interconnection of electricity and hydrogen, are an important measure for the consumption of intermittent renewable energy generation and the achievement of dual-carbon goals. Meanwhile, long-period hydrogen storage systems have attracted widespread attention due to their ability to compensate for medium- and long-term fluctuations in renewable energy. However, risk defense planning models involving electric-hydrogen energy systems are usually limited by model complexity as they must cover the complete charging and discharging cycles of long-cycle hydrogen storage. Therefore, this paper proposes a risk defense planning method for electric-hydrogen energy systems based on adaptive time-series aggregation, in which, firstly, an adaptive time-series aggregation algorithm is designed based on the predicted values of renewable energy sources and load demand to reduce the variable dimensions of the operational simulation; secondly, conditional value-at-risk theory is introduced to model renewable energy sources and electric load fluctuations using multivariate Gaussian distributions, which is used to quantify the potential operational risks of the system; Finally, taking into account the system investment and operation cost, an optimal planning model for electric-hydrogen energy system based on adaptive time-series aggregation is constructed with the optimization objective of minimizing the total annualized cost. Example results show that the proposed method can simultaneously consider the adjustment ability of long/short cycle hydrogen storage and the potential operating risks, improve the risk defense ability of the system, and effectively balance the system"s potential operating risks and investment costs.
Electric-hydrogen energy systems, which enable the interconnection of electricity and hydrogen, are an important measure for the consumption of intermittent renewable energy generation and the achievement of dual-carbon goals. Meanwhile, long-period hydrogen storage systems have attracted widespread attention due to their ability to compensate for medium- and long-term fluctuations in renewable energy. However, risk defense planning models involving electric-hydrogen energy systems are usually limited by model complexity as they must cover the complete charging and discharging cycles of long-cycle hydrogen storage. Therefore, this paper proposes a risk defense planning method for electric-hydrogen energy systems based on adaptive time-series aggregation, in which, firstly, an adaptive time-series aggregation algorithm is designed based on the predicted values of renewable energy sources and load demand to reduce the variable dimensions of the operational simulation; secondly, conditional value-at-risk theory is introduced to model renewable energy sources and electric load fluctuations using multivariate Gaussian distributions, which is used to quantify the potential operational risks of the system; Finally, taking into account the system investment and operation cost, an optimal planning model for electric-hydrogen energy system based on adaptive time-series aggregation is constructed with the optimization objective of minimizing the total annualized cost. Example results show that the proposed method can simultaneously consider the adjustment ability of long/short cycle hydrogen storage and the potential operating risks, improve the risk defense ability of the system, and effectively balance the system"s potential operating risks and investment costs.
Abstract:
With the rapid development of Internet of Vehicles technology, edge computing plays an increasingly important role in information processing applications of Internet of Vehicles. However, the computing resources and storage capa-bilities of vehicle equipment in the Internet of Vehicles are limited, and vehicles need to process large amounts of data and complex computing tasks. How to efficiently utilize edge computing resources is an important issue. A single offloading method may not be able to meet the needs of the scenario. Consider the collaboration of multiple offload-ing methods to improve the overall performance and flexibility of the system. In response to the above challenges, a multi-mode collaborative vehicle network edge computing task offloading solution based on deep reinforcement learning is proposed. By using Vehicle to Vehicle (V2V), Vehicle to Vehicle Alliance (V2A), Vehicle to Roadside Unit (V2R) and Roadside Unit to Base (Roadside Unit to Base) Station, R2B) and other communication methods realize collaborative processing of tasks with high computing density and high latency requirements. In order to cope with the complexity of the dynamic network environment, the Markov Decision Process (MDP) is used for problem modeling and optimization, and deep reinforcement learning is introduced to handle continuous actions and state space. In particular, the MCTO algorithm is proposed, which can efficiently adapt to the dynamic environment of the Internet of Vehicles and significantly reduce the delay of the entire Internet of Vehicles system. The simulation results show that the proposed MCTO algorithm has good convergence and is significantly improved in terms of system delay compared with other reinforcement learning algorithms, with the overall performance improved by 28.67%.
With the rapid development of Internet of Vehicles technology, edge computing plays an increasingly important role in information processing applications of Internet of Vehicles. However, the computing resources and storage capa-bilities of vehicle equipment in the Internet of Vehicles are limited, and vehicles need to process large amounts of data and complex computing tasks. How to efficiently utilize edge computing resources is an important issue. A single offloading method may not be able to meet the needs of the scenario. Consider the collaboration of multiple offload-ing methods to improve the overall performance and flexibility of the system. In response to the above challenges, a multi-mode collaborative vehicle network edge computing task offloading solution based on deep reinforcement learning is proposed. By using Vehicle to Vehicle (V2V), Vehicle to Vehicle Alliance (V2A), Vehicle to Roadside Unit (V2R) and Roadside Unit to Base (Roadside Unit to Base) Station, R2B) and other communication methods realize collaborative processing of tasks with high computing density and high latency requirements. In order to cope with the complexity of the dynamic network environment, the Markov Decision Process (MDP) is used for problem modeling and optimization, and deep reinforcement learning is introduced to handle continuous actions and state space. In particular, the MCTO algorithm is proposed, which can efficiently adapt to the dynamic environment of the Internet of Vehicles and significantly reduce the delay of the entire Internet of Vehicles system. The simulation results show that the proposed MCTO algorithm has good convergence and is significantly improved in terms of system delay compared with other reinforcement learning algorithms, with the overall performance improved by 28.67%.
Abstract:
In recent years, image steganalysis algorithms have developed rapidly, and a series of advanced deep learning-based steganalysis models have been continuously proposed. However, existing steganalysis algorithms are unable to achieve precise weight allocation in areas with complex and smooth image textures, hindering their further development. To address the instability in feature extraction and analysis of steganalysis algorithms, this paper proposes a image steganalysis algorithm based on complexity matching and attention mechanism. First, using the characteristics of steganography algorithm tends to embed in texture complex region, designing complexity matching strategy, independently extracting texture complex block and texture smooth block features, aggregating strong steganographic signal region, and improving the extraction ability of the model for weak steganographic signals. Second, through the design of the improved convolutional attention mechanism, the attention to different image regions is effectively and reasonably allocated to improve the network for the important features and texture regions. The experimental results show that the proposed algorithm improves the steganalysis performance of several steganographic algorithms on two public datasets, BOSSBase v1.01 and Alaska2, compared with the existing models.
In recent years, image steganalysis algorithms have developed rapidly, and a series of advanced deep learning-based steganalysis models have been continuously proposed. However, existing steganalysis algorithms are unable to achieve precise weight allocation in areas with complex and smooth image textures, hindering their further development. To address the instability in feature extraction and analysis of steganalysis algorithms, this paper proposes a image steganalysis algorithm based on complexity matching and attention mechanism. First, using the characteristics of steganography algorithm tends to embed in texture complex region, designing complexity matching strategy, independently extracting texture complex block and texture smooth block features, aggregating strong steganographic signal region, and improving the extraction ability of the model for weak steganographic signals. Second, through the design of the improved convolutional attention mechanism, the attention to different image regions is effectively and reasonably allocated to improve the network for the important features and texture regions. The experimental results show that the proposed algorithm improves the steganalysis performance of several steganographic algorithms on two public datasets, BOSSBase v1.01 and Alaska2, compared with the existing models.
Abstract:
Small objects in images often present significant challenges during segmentation due to their irregular shapes and blurred boundaries. These challenges primarily include difficulties in feature extraction, loss of edge details, and significant noise interference. To effectively address these challenges, we propose an efficient small object segmentation model named LESO-Net, a lightweight and efficient object segmentation Network for small objects, based on You Only Look Once (YOLO) v8n-seg. Initially, we integrate a Large Separable Convolution Attention (LSKA) module into the Neck network, which not only enhances segmentation accuracy but also reduces computational complexity and memory usage. In addition, to specifically address the unstable shapes of small objects, we replace the C2f module in the backbone model with our improved Deformable Convolutional Networks (DCNv2). This modification significantly enhances feature extraction and adaptive generalization capabilities for small objects of varying shapes. Furthermore, to further improve the model's performance, we ameliorate the loss function, thereby effectively tackling class imbalances and insufficient bounding box accuracy. We validated the effectiveness of LESO-Net on a self-constructed bubble dataset and a public high-resolution SAR images dataset (HRSID). Compared to the original version of YOLOv8n-seg, LESO-Net achieved a precision improvement of 1.2 percentage points and 2.5 percentage points, along with average precision enhancements of 0.2 percentage point and 1.2 percentage points, respectively, while the number of parameters was reduced by nearly 10%. This result effectively demonstrates the superior performance of the LESO-Net model, meeting the accuracy requirements for segmenting small targets in practical remote sensing scenarios.
Small objects in images often present significant challenges during segmentation due to their irregular shapes and blurred boundaries. These challenges primarily include difficulties in feature extraction, loss of edge details, and significant noise interference. To effectively address these challenges, we propose an efficient small object segmentation model named LESO-Net, a lightweight and efficient object segmentation Network for small objects, based on You Only Look Once (YOLO) v8n-seg. Initially, we integrate a Large Separable Convolution Attention (LSKA) module into the Neck network, which not only enhances segmentation accuracy but also reduces computational complexity and memory usage. In addition, to specifically address the unstable shapes of small objects, we replace the C2f module in the backbone model with our improved Deformable Convolutional Networks (DCNv2). This modification significantly enhances feature extraction and adaptive generalization capabilities for small objects of varying shapes. Furthermore, to further improve the model's performance, we ameliorate the loss function, thereby effectively tackling class imbalances and insufficient bounding box accuracy. We validated the effectiveness of LESO-Net on a self-constructed bubble dataset and a public high-resolution SAR images dataset (HRSID). Compared to the original version of YOLOv8n-seg, LESO-Net achieved a precision improvement of 1.2 percentage points and 2.5 percentage points, along with average precision enhancements of 0.2 percentage point and 1.2 percentage points, respectively, while the number of parameters was reduced by nearly 10%. This result effectively demonstrates the superior performance of the LESO-Net model, meeting the accuracy requirements for segmenting small targets in practical remote sensing scenarios.
Abstract:
In the financial market, gold futures prices are influenced by a variety of factors, and accurate prediction of these prices holds significant importance. To address this issue, a new model for predicting gold futures prices has been proposed that integrates multiple data sources, combining a LightGBM(Light Gradient Boosting Machine)feature selection method with an LSTM model. Firstly, the paper preprocesses the acquired macroeconomic indicators and technical indicators. It then annotates the sentiment tendencies of unstructured news headlines using various methods, leading to the construction of a weighted sentiment index. Additionally, it merges the search indices of multiple keywords from Baidu into a comprehensive Baidu search index. Secondly, the LightGBM method is used to rank the importance of features from macroeconomic and technical indicators, extracting key features. Finally, the selected features, along with the weighted sentiment index and the comprehensive Baidu search index, serve as input variables for the LSTM forecasting model. Empirical results show that the LightGBM-LSTM model with multi-source data has excellent prediction performance and the model prediction error is the smallest. Compared with the benchmark model, it can make a more accurate prediction of the closing price of gold futures.
In the financial market, gold futures prices are influenced by a variety of factors, and accurate prediction of these prices holds significant importance. To address this issue, a new model for predicting gold futures prices has been proposed that integrates multiple data sources, combining a LightGBM(Light Gradient Boosting Machine)feature selection method with an LSTM model. Firstly, the paper preprocesses the acquired macroeconomic indicators and technical indicators. It then annotates the sentiment tendencies of unstructured news headlines using various methods, leading to the construction of a weighted sentiment index. Additionally, it merges the search indices of multiple keywords from Baidu into a comprehensive Baidu search index. Secondly, the LightGBM method is used to rank the importance of features from macroeconomic and technical indicators, extracting key features. Finally, the selected features, along with the weighted sentiment index and the comprehensive Baidu search index, serve as input variables for the LSTM forecasting model. Empirical results show that the LightGBM-LSTM model with multi-source data has excellent prediction performance and the model prediction error is the smallest. Compared with the benchmark model, it can make a more accurate prediction of the closing price of gold futures.
Abstract:
To significantly enhance the economic performance of electric vehicles operating in platoons and mitigate the rate of power battery capacity degradation, a comprehensive study on driving speed planning was undertaken specifically for a given electric truck platoon. Firstly, a single vehicle energy consumption model for electric trucks was established; then, leveraging this foundational single-vehicle model, a strategy incorporating a fixed headway following distance, along with aerodynamic drag coefficients derived through precise fitting for each vehicle within the platoon at varying distances, facilitated the development of an energy consumption model tailored specifically for the platoon operation. Subsequently, the alternating direction method of multipliers (ADMM) was used to optimize the speed of the vehicle platoon, aiming to minimize energy consumption while adhering to constraints such as travel time and road speed limits. Ultimately, the influence of the planned speed on both the average energy consumption of the platoon and the battery capacity decay was simulated and analyzed. Research has shown that compared to the speed of adaptive cruise control and the speed planned by the quadratic programming method based on road slope acceleration in the platoon, driving at the speed planned by ADMM in the platoon the electric vehicle can save energy by 9.30% and 7.76%, respectively; and with the increase of driving mileage, the energy-saving effect of driving at the speed planned by ADMM becomes more and more obvious; After one year of operation in the vehicle platoon, the battery capacity decay would be decreased by 9.69% and 2.54%, indicating that driving at the speed planned by ADMM can slow down the aging rate of the battery, and over time, the difference in battery capacity decay between these three driving speeds becomes increasingly significant.
To significantly enhance the economic performance of electric vehicles operating in platoons and mitigate the rate of power battery capacity degradation, a comprehensive study on driving speed planning was undertaken specifically for a given electric truck platoon. Firstly, a single vehicle energy consumption model for electric trucks was established; then, leveraging this foundational single-vehicle model, a strategy incorporating a fixed headway following distance, along with aerodynamic drag coefficients derived through precise fitting for each vehicle within the platoon at varying distances, facilitated the development of an energy consumption model tailored specifically for the platoon operation. Subsequently, the alternating direction method of multipliers (ADMM) was used to optimize the speed of the vehicle platoon, aiming to minimize energy consumption while adhering to constraints such as travel time and road speed limits. Ultimately, the influence of the planned speed on both the average energy consumption of the platoon and the battery capacity decay was simulated and analyzed. Research has shown that compared to the speed of adaptive cruise control and the speed planned by the quadratic programming method based on road slope acceleration in the platoon, driving at the speed planned by ADMM in the platoon the electric vehicle can save energy by 9.30% and 7.76%, respectively; and with the increase of driving mileage, the energy-saving effect of driving at the speed planned by ADMM becomes more and more obvious; After one year of operation in the vehicle platoon, the battery capacity decay would be decreased by 9.69% and 2.54%, indicating that driving at the speed planned by ADMM can slow down the aging rate of the battery, and over time, the difference in battery capacity decay between these three driving speeds becomes increasingly significant.
Abstract:
An effective detection scheme based on decoding electroencephalogram (EEG) signals under different states of sickness is proposed, which is helpful for studying methods to alleviate virtual reality motion sickness (VRMS). This article uses multivariate variational mode decomposition (MVMD) to divide EEG into five frequency bands and divides the data into different sickness state groups based on the results of the motion sickness scale. The phase-locking value (PLV) method is used to calculate the functional connections within and between EEG frequency bands to construct a super adjacent matrix, and classification recognition is performed based on models such as support vector machine (SVM) and convolutional neural network (CNN). The research results show that the fusion of three topological features with significant differences in clustering coefficient, local efficiency, and weighted node degree results in sickness vs Non sickness, high sickness vs The highest average classification accuracy in the two tasks of low sickness was 91.70% and 96.00%, respectively. In addition, this article also directly inputs the super adjacency matrix into the CNN model, achieving average classification accuracies of 93.40% and 98.50% in two tasks, respectively. The results indicate that the method proposed in this study can be used for the detection of motion sickness in virtual reality and provide reference for further research on the impact of motion sickness on the functional coupling of various brain regions.
An effective detection scheme based on decoding electroencephalogram (EEG) signals under different states of sickness is proposed, which is helpful for studying methods to alleviate virtual reality motion sickness (VRMS). This article uses multivariate variational mode decomposition (MVMD) to divide EEG into five frequency bands and divides the data into different sickness state groups based on the results of the motion sickness scale. The phase-locking value (PLV) method is used to calculate the functional connections within and between EEG frequency bands to construct a super adjacent matrix, and classification recognition is performed based on models such as support vector machine (SVM) and convolutional neural network (CNN). The research results show that the fusion of three topological features with significant differences in clustering coefficient, local efficiency, and weighted node degree results in sickness vs Non sickness, high sickness vs The highest average classification accuracy in the two tasks of low sickness was 91.70% and 96.00%, respectively. In addition, this article also directly inputs the super adjacency matrix into the CNN model, achieving average classification accuracies of 93.40% and 98.50% in two tasks, respectively. The results indicate that the method proposed in this study can be used for the detection of motion sickness in virtual reality and provide reference for further research on the impact of motion sickness on the functional coupling of various brain regions.
Abstract:
Aiming at the shortcomings of traditional graph neural network methods in point cloud semantic segmentation, such as high supervision accuracy requirement, node label transfer can only be unidirectional, and global information is not taken into account, this paper proposes a point cloud semantic segmentation method based on bidirectional attention mechanism. The algorithm segments the point cloud into superpoints and constructs a superpoint graph thus introducing the point cloud classification problem into the superpoint graph framework. After that, using the two-way attention module, it alternately pays attention to the superpoints and updates the superpoint features according to the weights of the neighboring superpoints, so as to achieve the two-way transfer of information. Also compared to previous graph pooling methods, this paper applies both maximum pooling and average pooling and combines the pooled features. In this paper, we use the public dataset Semantic3D for training and experiments, and the results show that the proposed method can effectively correct the annotation error and combine the long-range information, and the mIoU and oAcc of the dataset are 75.4% and 95.1%, respectively, which reflect better label delivery mechanism and higher classification accuracy compared with state of art methods.
Aiming at the shortcomings of traditional graph neural network methods in point cloud semantic segmentation, such as high supervision accuracy requirement, node label transfer can only be unidirectional, and global information is not taken into account, this paper proposes a point cloud semantic segmentation method based on bidirectional attention mechanism. The algorithm segments the point cloud into superpoints and constructs a superpoint graph thus introducing the point cloud classification problem into the superpoint graph framework. After that, using the two-way attention module, it alternately pays attention to the superpoints and updates the superpoint features according to the weights of the neighboring superpoints, so as to achieve the two-way transfer of information. Also compared to previous graph pooling methods, this paper applies both maximum pooling and average pooling and combines the pooled features. In this paper, we use the public dataset Semantic3D for training and experiments, and the results show that the proposed method can effectively correct the annotation error and combine the long-range information, and the mIoU and oAcc of the dataset are 75.4% and 95.1%, respectively, which reflect better label delivery mechanism and higher classification accuracy compared with state of art methods.
Abstract:
Knee osteoarthritis is often misdiagnosed due to unclear categorization during examination, which affects the therapeutic effect, and existing detection methods perform poorly in terms of detection accuracy and accurate differentiation of lesion categories. To address the above problems this paper proposes an improved classification algorithm based on YOLOv8 for the detection of osteoarthritis of the knee joint, aiming to improve the accuracy of detection and classification. First, the distribution mapping module (V-PENet) is designed to help YOLO network to be trained better by preprocessing the input images and mapping the complex data distributions, uniformly, into simple data distributions. Meanwhile, a semantic perception module (PTB) is added to enhance the model's context-awareness ability and its understanding of global information. Secondly, the CA coordinate attention mechanism is introduced to enrich the feature information and further enhance the model's ability to capture mesoscale target information. Finally, the WIoUv3 loss function is used to replace the original CIoU loss function to optimize the positioning accuracy. Compared with the benchmark model YOLOv8n, the accuracy is improved by 9.9%, mAP@0.5 reaches 81.2% and mAP0.5:0.95 reaches 64.3%. Compared with other detection methods, the improved model proposed in this paper has a significant advantage in accuracy and can better meet the needs of knee osteoarthritis detection and classification.
Knee osteoarthritis is often misdiagnosed due to unclear categorization during examination, which affects the therapeutic effect, and existing detection methods perform poorly in terms of detection accuracy and accurate differentiation of lesion categories. To address the above problems this paper proposes an improved classification algorithm based on YOLOv8 for the detection of osteoarthritis of the knee joint, aiming to improve the accuracy of detection and classification. First, the distribution mapping module (V-PENet) is designed to help YOLO network to be trained better by preprocessing the input images and mapping the complex data distributions, uniformly, into simple data distributions. Meanwhile, a semantic perception module (PTB) is added to enhance the model's context-awareness ability and its understanding of global information. Secondly, the CA coordinate attention mechanism is introduced to enrich the feature information and further enhance the model's ability to capture mesoscale target information. Finally, the WIoUv3 loss function is used to replace the original CIoU loss function to optimize the positioning accuracy. Compared with the benchmark model YOLOv8n, the accuracy is improved by 9.9%, mAP@0.5 reaches 81.2% and mAP0.5:0.95 reaches 64.3%. Compared with other detection methods, the improved model proposed in this paper has a significant advantage in accuracy and can better meet the needs of knee osteoarthritis detection and classification.
Abstract:
Transmission lines are critical elements for the stable operation of power systems. When a fault occurs on a transmission line, the ability to quickly and accurately identify the cause of the fault is of great significance for the safe and stable operation of the power system. To address the low accuracy problem present in existing transmission line fault cause identification schemes, this paper proposes a method for identifying transmission line fault causes based on multi-source data fusion. First, external interference factors and waveform characteristics corresponding to different types of transmission line faults are analyzed to provide theoretical support for multi-source data input. Next, the Gramian Angular Field and feature encoding techniques are employed to preprocess fault information, constructing feature representations for different fault types from the perspectives of time-series waveforms, two-dimensional images, and discrete features. Furthermore, a method is designed that fuses adaptive boundary parameters with Long Short-Term Memory neural networks, Convolutional Neural Networks , and Artificial Neural Networks to classify and identify the causes of transmission line faults. Finally, the effectiveness and superiority of the proposed method are verified through comparative tests on real-world data, providing a reliable solution for accomplishing the high-precision task of transmission line fault cause identification.
Transmission lines are critical elements for the stable operation of power systems. When a fault occurs on a transmission line, the ability to quickly and accurately identify the cause of the fault is of great significance for the safe and stable operation of the power system. To address the low accuracy problem present in existing transmission line fault cause identification schemes, this paper proposes a method for identifying transmission line fault causes based on multi-source data fusion. First, external interference factors and waveform characteristics corresponding to different types of transmission line faults are analyzed to provide theoretical support for multi-source data input. Next, the Gramian Angular Field and feature encoding techniques are employed to preprocess fault information, constructing feature representations for different fault types from the perspectives of time-series waveforms, two-dimensional images, and discrete features. Furthermore, a method is designed that fuses adaptive boundary parameters with Long Short-Term Memory neural networks, Convolutional Neural Networks , and Artificial Neural Networks to classify and identify the causes of transmission line faults. Finally, the effectiveness and superiority of the proposed method are verified through comparative tests on real-world data, providing a reliable solution for accomplishing the high-precision task of transmission line fault cause identification.
Abstract:
For the quadrotor with model uncertainty and quantised input, this paper proposes a new control strategy by combining fixed-time command filtering with backstepping under quantised input. Firstly, the virtual control signal in the traditional backstepping method is used as the input of fixed-time command filtering, so that the derivative of the virtual control signal can be obtained directly, which effectively avoids the problem of ‘differential explosion’ brought by the traditional backstepping method, and the filtering error compensation signal is designed to fill the defect that the filtering error affects the system performance. Then, an adaptive law for the quantisation parameters is designed, which can still be applied when the a priori information of the quantisation parameters is uncertain. Then, it is verified that all the signals in the closed-loop system are bounded at fixed time, and the position and attitude tracking errors of the quadrotor can reach a sufficiently small region of the origin at fixed time. Finally, simulation experiments are given to verify the feasibility and superiority of the control method.
For the quadrotor with model uncertainty and quantised input, this paper proposes a new control strategy by combining fixed-time command filtering with backstepping under quantised input. Firstly, the virtual control signal in the traditional backstepping method is used as the input of fixed-time command filtering, so that the derivative of the virtual control signal can be obtained directly, which effectively avoids the problem of ‘differential explosion’ brought by the traditional backstepping method, and the filtering error compensation signal is designed to fill the defect that the filtering error affects the system performance. Then, an adaptive law for the quantisation parameters is designed, which can still be applied when the a priori information of the quantisation parameters is uncertain. Then, it is verified that all the signals in the closed-loop system are bounded at fixed time, and the position and attitude tracking errors of the quadrotor can reach a sufficiently small region of the origin at fixed time. Finally, simulation experiments are given to verify the feasibility and superiority of the control method.
Abstract:
With the acceleration of the pace of global energy transformation, the proportion of renewable energy generation continues to increase, and the volatility and uncertainty of the power system also increase, which puts higher requirements on the flexible adjustment ability and stable operation of the power system. Therefore, this paper reviews the key role of energy storage system (ESS) in assisting the operation control of new power systems and its optimal configuration technology. Firstly, the application scenarios of the energy storage system in the aspects of peak regulation, frequency modulation and power quality improvement are systematically reviewed, and the optimal configuration strategy of the energy storage system under these application scenarios is deeply discussed. Secondly, the economy, reliability and stability of energy storage optimization are analyzed and studied in detail. Then, the advantages, disadvantages and applicability of different optimization algorithms (including linear programming, dynamic programming, genetic algorithm, particle swarm optimization, etc.) are compared. Finally, the challenges and future development direction of energy storage optimization configuration technology are revealed. The analysis shows that scientific and reasonable allocation of energy storage resources is crucial to improve the operating efficiency, stability and reliability of the power system, and is a key technical approach to promote the wide application of renewable energy and achieve the goal of smart grid construction. In the future, with the development of new composite energy storage technology and the introduction of intelligent optimization algorithms, energy storage optimization configuration technology is expected to play a more prominent role in the sustainable development of new power systems.
With the acceleration of the pace of global energy transformation, the proportion of renewable energy generation continues to increase, and the volatility and uncertainty of the power system also increase, which puts higher requirements on the flexible adjustment ability and stable operation of the power system. Therefore, this paper reviews the key role of energy storage system (ESS) in assisting the operation control of new power systems and its optimal configuration technology. Firstly, the application scenarios of the energy storage system in the aspects of peak regulation, frequency modulation and power quality improvement are systematically reviewed, and the optimal configuration strategy of the energy storage system under these application scenarios is deeply discussed. Secondly, the economy, reliability and stability of energy storage optimization are analyzed and studied in detail. Then, the advantages, disadvantages and applicability of different optimization algorithms (including linear programming, dynamic programming, genetic algorithm, particle swarm optimization, etc.) are compared. Finally, the challenges and future development direction of energy storage optimization configuration technology are revealed. The analysis shows that scientific and reasonable allocation of energy storage resources is crucial to improve the operating efficiency, stability and reliability of the power system, and is a key technical approach to promote the wide application of renewable energy and achieve the goal of smart grid construction. In the future, with the development of new composite energy storage technology and the introduction of intelligent optimization algorithms, energy storage optimization configuration technology is expected to play a more prominent role in the sustainable development of new power systems.
Abstract:
To address the problem that existing algorithms cannot accurately perform the detection task of tire side pressure marking text area due to interference from dark background, this paper proposes a tire side pressure marking text area detection algorithm TireYOLO based on the improved YOLOv8s. Firstly, a lightweight feature extraction module RepNCSPELAN4_CAA is proposed to replace the C2f module in the main body network of YOLOv8s, which can capture contextual feature information better through the CAA attention mechanism and improve model accuracy while reducing computational cost. Secondly, in order to better fuse features while keeping lightweight, a dynamic sampling lightweight large receptive field feature fusion neck structure DyLKVHSPAN is adopted, and an efficient feature fusion network HSPAN is proposed by improving the HSFPN network structure. At the same time, a large receptive field feature fusion module UniLKVoVCSP is proposed to dynamically sample features through the Dysample upsampling operator, so that DyLKVHSPAN can fuse features better with large receptive field while keeping lightweight. Finally, a low computational cost detection head DWLHead is proposed, which redesigns the structure of the original detection head to improve detection performance while reducing computational cost. Experimental results show that compared with YOLOv8s, the TireYOLO algorithm improves the mAP0.5:0.95 of the key information area detection in tire side by 2.5 percentage points, reduces the number of parameters by 48.4%, and reduces the computational cost by 49.6%, verifying the superiority of the proposed algorithm in detecting tire side text area.
To address the problem that existing algorithms cannot accurately perform the detection task of tire side pressure marking text area due to interference from dark background, this paper proposes a tire side pressure marking text area detection algorithm TireYOLO based on the improved YOLOv8s. Firstly, a lightweight feature extraction module RepNCSPELAN4_CAA is proposed to replace the C2f module in the main body network of YOLOv8s, which can capture contextual feature information better through the CAA attention mechanism and improve model accuracy while reducing computational cost. Secondly, in order to better fuse features while keeping lightweight, a dynamic sampling lightweight large receptive field feature fusion neck structure DyLKVHSPAN is adopted, and an efficient feature fusion network HSPAN is proposed by improving the HSFPN network structure. At the same time, a large receptive field feature fusion module UniLKVoVCSP is proposed to dynamically sample features through the Dysample upsampling operator, so that DyLKVHSPAN can fuse features better with large receptive field while keeping lightweight. Finally, a low computational cost detection head DWLHead is proposed, which redesigns the structure of the original detection head to improve detection performance while reducing computational cost. Experimental results show that compared with YOLOv8s, the TireYOLO algorithm improves the mAP0.5:0.95 of the key information area detection in tire side by 2.5 percentage points, reduces the number of parameters by 48.4%, and reduces the computational cost by 49.6%, verifying the superiority of the proposed algorithm in detecting tire side text area.
Abstract:
Absrtact: The open sharing of remote sensing data is mainly ensured through data policies and data sharing service platforms. Data policies determine which data can be shared, while sharing platforms provide tools for data retrieval and download. In recent years, China has continued to make efforts in formulating data policies and building sharing platforms, greatly promoting the application of domestic civil remote sensing satellite data represented by high-resolution projects and space infrastructure. With the continuous expansion of remote sensing business applications, the demand for domestic commercial remote sensing satellite data is increasing, and it is urgent to build a one-stop service platform for civil and commercial remote sensing data. This article explores the current status of open sharing service platforms for remote sensing data at home and abroad, analyzes the retrieval functions of major commercial remote sensing data sales platforms at home and abroad, systematically sorts out the requirements for one-stop service platforms for civil and commercial remote sensing data, designs a technical process, and finally combines with the land observation satellite data service platform to provide optimized retrieval functions for public welfare needs, and proposes a design idea for adding a commercial remote sensing data retrieval module.
Absrtact: The open sharing of remote sensing data is mainly ensured through data policies and data sharing service platforms. Data policies determine which data can be shared, while sharing platforms provide tools for data retrieval and download. In recent years, China has continued to make efforts in formulating data policies and building sharing platforms, greatly promoting the application of domestic civil remote sensing satellite data represented by high-resolution projects and space infrastructure. With the continuous expansion of remote sensing business applications, the demand for domestic commercial remote sensing satellite data is increasing, and it is urgent to build a one-stop service platform for civil and commercial remote sensing data. This article explores the current status of open sharing service platforms for remote sensing data at home and abroad, analyzes the retrieval functions of major commercial remote sensing data sales platforms at home and abroad, systematically sorts out the requirements for one-stop service platforms for civil and commercial remote sensing data, designs a technical process, and finally combines with the land observation satellite data service platform to provide optimized retrieval functions for public welfare needs, and proposes a design idea for adding a commercial remote sensing data retrieval module.
Abstract:
To tackle the current challenges of low efficiency, subpar performance, and inadequate real-time capabilities in bridge crack detection, this paper introduces an improved YOLOv8 model-based drone image detection method for bridge cracks. Firstly, the dynamic snake convolution kernel is integrated into the C2f module in the backbone of YOLOv8 to enhance the crack feature extraction ability. Then, the CAM module was introduced to improve the ability of small target detection. Finally, the influence of low-quality data sets on detection results was reduced by optimizing the prediction box loss function. The experimental results show that the GFOLPs of the improved model reaches 14.4 and mAP@50 reaches 94%. Compared with the basic model, the accuracy of the improved model is greatly improved, and the detection speed reaches 147 frames /s, which can meet the requirements of real-time crack detection by UAVs.
To tackle the current challenges of low efficiency, subpar performance, and inadequate real-time capabilities in bridge crack detection, this paper introduces an improved YOLOv8 model-based drone image detection method for bridge cracks. Firstly, the dynamic snake convolution kernel is integrated into the C2f module in the backbone of YOLOv8 to enhance the crack feature extraction ability. Then, the CAM module was introduced to improve the ability of small target detection. Finally, the influence of low-quality data sets on detection results was reduced by optimizing the prediction box loss function. The experimental results show that the GFOLPs of the improved model reaches 14.4 and mAP@50 reaches 94%. Compared with the basic model, the accuracy of the improved model is greatly improved, and the detection speed reaches 147 frames /s, which can meet the requirements of real-time crack detection by UAVs.
Abstract:
To accurately detect road damages with large size differences and small scales in Vehicle-mounted-images, this paper presents a real-time Vehicle-mounted-images road damage detection model based on improved YOLOv5s, termed as VRD-YOLO(Vehicle-mounted-images Road Damage Detection -YOLO). Firstly, a Channel Mix Slide Transformer module is proposed, which enhances the model's global context modeling capability and strengthens the extraction of fine-grained road lesion semantic feature information. Secondly, a generalized feature pyramid with cross-layer fusion and cross-scale fusion is introduced to expand the network sensing field and strengthen the fusion of multi-scale damage features. Thirdly, to optimize the model feature response and improve the model’s detection performance, a dynamic detection head is designed to achieve scale perception, spatial perception, and task perception. Finally, the Vehicle-mounted Images Road Damage Dataset (VIRDD) was constructed to expand the number and type of existing road damage datasets, and ablation and comparison experiments were conducted based on this dataset. The experimental results show that VRD-YOLO achieves a detection accuracy of 74.45% mAP@0.5 on the VIRDD dataset, with a detection speed reaching 28.56 FPS. Compared to the YOLOv5s model, VRD-YOLO improves the precision, recall, F1 score, and mean average precision by 2.79, 2.32, 2.54, and 3.19 percentage points, respectively. Additionally, compared with six other classical and novel object detection models, VRD-YOLO attains the highest detection accuracy with the smallest model parameter count of 9.68 million, verifying the effectiveness and superiority of the proposed method.
To accurately detect road damages with large size differences and small scales in Vehicle-mounted-images, this paper presents a real-time Vehicle-mounted-images road damage detection model based on improved YOLOv5s, termed as VRD-YOLO(Vehicle-mounted-images Road Damage Detection -YOLO). Firstly, a Channel Mix Slide Transformer module is proposed, which enhances the model's global context modeling capability and strengthens the extraction of fine-grained road lesion semantic feature information. Secondly, a generalized feature pyramid with cross-layer fusion and cross-scale fusion is introduced to expand the network sensing field and strengthen the fusion of multi-scale damage features. Thirdly, to optimize the model feature response and improve the model’s detection performance, a dynamic detection head is designed to achieve scale perception, spatial perception, and task perception. Finally, the Vehicle-mounted Images Road Damage Dataset (VIRDD) was constructed to expand the number and type of existing road damage datasets, and ablation and comparison experiments were conducted based on this dataset. The experimental results show that VRD-YOLO achieves a detection accuracy of 74.45% mAP@0.5 on the VIRDD dataset, with a detection speed reaching 28.56 FPS. Compared to the YOLOv5s model, VRD-YOLO improves the precision, recall, F1 score, and mean average precision by 2.79, 2.32, 2.54, and 3.19 percentage points, respectively. Additionally, compared with six other classical and novel object detection models, VRD-YOLO attains the highest detection accuracy with the smallest model parameter count of 9.68 million, verifying the effectiveness and superiority of the proposed method.
Abstract:
Addressing the issues of poor real-time performance and insufficient predictive capability in continuous gesture recognition tasks based on surface electromyography (sEMG), a method utilizing Gaussian Mixture Model-Hidden Markov Models (GMM-HMMs) and Viterbi traceback is proposed for continuous gesture recognition. This approach leverages Gaussian Mixture Models-Hidden Markov Models (GMM-HMMs) to classify hand gestures into idle, ascending, steady, and descending states. A refined Viterbi sliding window marginalization technique is implemented to ensure prolonged connections between adjacent windows, enabling anticipatory prediction of subsequent gesture states. Moreover, a dynamic threshold model based on maximum likelihood is incorporated to accurately categorize gestures. In a study where 8 participants performed 12 continuous dual-gesture tasks, the method proposed attained an average recognition rate of 98.1% with a prediction time of 71ms, surpassing the LSTM model (94.2%, 309ms) and the GRU model (93.8%, 300ms), resulting in a considerable decrease in prediction time.
Addressing the issues of poor real-time performance and insufficient predictive capability in continuous gesture recognition tasks based on surface electromyography (sEMG), a method utilizing Gaussian Mixture Model-Hidden Markov Models (GMM-HMMs) and Viterbi traceback is proposed for continuous gesture recognition. This approach leverages Gaussian Mixture Models-Hidden Markov Models (GMM-HMMs) to classify hand gestures into idle, ascending, steady, and descending states. A refined Viterbi sliding window marginalization technique is implemented to ensure prolonged connections between adjacent windows, enabling anticipatory prediction of subsequent gesture states. Moreover, a dynamic threshold model based on maximum likelihood is incorporated to accurately categorize gestures. In a study where 8 participants performed 12 continuous dual-gesture tasks, the method proposed attained an average recognition rate of 98.1% with a prediction time of 71ms, surpassing the LSTM model (94.2%, 309ms) and the GRU model (93.8%, 300ms), resulting in a considerable decrease in prediction time.
Abstract:
Abstract: To enhance the accuracy of neural network models in predicting the remaining useful life of lithium-ion batteries, this paper proposed a prediction method based on the Attention-GRU (Gated Recurrent Unit) model. Firstly, based on the Random Walk battery data publicly available from NASA, health factors that significantly correlated with battery capacity decay were extracted from direct measurement data such as battery voltage, current, and time, and the correlation between these health factors and battery capacity was calculated. Secondly, an Attention-GRU model was constructed to learn the patterns of change in health factors, allocate attention weights based on correlation, obtain attention vectors, and adjust the output of the hidden layer. Finally, predictions for the remaining useful life of the battery were made, and predictions for the B05 series of batteries were added to explore the model's generalization capabilities. Experimental results indicated that when using the Attention-GRU model to predict the Random Walk batteries, the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) were around 0.015, and for the B05 series batteries, the MAE and RMSE were below 0.01. The prediction accuracy was superior to comparative methods, demonstrating high precision and good generalization ability.
Abstract: To enhance the accuracy of neural network models in predicting the remaining useful life of lithium-ion batteries, this paper proposed a prediction method based on the Attention-GRU (Gated Recurrent Unit) model. Firstly, based on the Random Walk battery data publicly available from NASA, health factors that significantly correlated with battery capacity decay were extracted from direct measurement data such as battery voltage, current, and time, and the correlation between these health factors and battery capacity was calculated. Secondly, an Attention-GRU model was constructed to learn the patterns of change in health factors, allocate attention weights based on correlation, obtain attention vectors, and adjust the output of the hidden layer. Finally, predictions for the remaining useful life of the battery were made, and predictions for the B05 series of batteries were added to explore the model's generalization capabilities. Experimental results indicated that when using the Attention-GRU model to predict the Random Walk batteries, the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) were around 0.015, and for the B05 series batteries, the MAE and RMSE were below 0.01. The prediction accuracy was superior to comparative methods, demonstrating high precision and good generalization ability.
Abstract:
With the development of multimedia services and mobile networks, in order to achieve the data interconnection and interoperability of smart cities, medium and high point surveillance equipment, Internet of Things (IoT), AI and other technologies make the data collection, uploading and organization of urban surveillance video automated, effectively providing active discovery, timely warning and early intervention of urban security risks. However, the huge data volume of surveillance video puts huge pressure on the transmission networks, for this reason, this paper proposes a neural network model image and video quality enhancement algorithm based on semantic feature extraction. The algorithm firstly proposes a joint optimization of the degradation model and then reconstruction model to address the problems of existing image and video recovery enhancement methods. The proposed model is validated on a publicly available dataset and compared with existing algorithms, the proposed method can achieve a 50% improvement in score compared to PULSE method and is close to the original HD image and video quality. In terms of user evaluation, 81% of the reconstruction results were found to be superior to the comparison algorithm. The results show that the proposed algorithm has higher reconstructed image and video quality.
With the development of multimedia services and mobile networks, in order to achieve the data interconnection and interoperability of smart cities, medium and high point surveillance equipment, Internet of Things (IoT), AI and other technologies make the data collection, uploading and organization of urban surveillance video automated, effectively providing active discovery, timely warning and early intervention of urban security risks. However, the huge data volume of surveillance video puts huge pressure on the transmission networks, for this reason, this paper proposes a neural network model image and video quality enhancement algorithm based on semantic feature extraction. The algorithm firstly proposes a joint optimization of the degradation model and then reconstruction model to address the problems of existing image and video recovery enhancement methods. The proposed model is validated on a publicly available dataset and compared with existing algorithms, the proposed method can achieve a 50% improvement in score compared to PULSE method and is close to the original HD image and video quality. In terms of user evaluation, 81% of the reconstruction results were found to be superior to the comparison algorithm. The results show that the proposed algorithm has higher reconstructed image and video quality.
Abstract:
In order to solve the problem of insufficient forecasting accuracy caused by many factors such as nonlinear and time series of load data, a short-term load forecasting model based on CEEMDAN and HBA-BiGRU-SelfAttention was proposed. Firstly, the random forest algorithm is used to extract the features of meteorological factors, which ensures the characteristics of data and reduces the complexity of data. Secondly, the adaptive noise complete set empirical mode decomposition algorithm is used to decompose the original load data, and some relatively stable modal components are obtained. Then, meteorological factors and modal components extracted by feature are taken as input data, and BiGRU-SelfAttention model is used for prediction. For the problem that it is difficult to select the optimal solution for hyperparameters of BiGRU-SelfAttention model, The Honey Badger algorithm is introduced to optimize the hyperparameters of BiGRU-SelfAttention model. Finally, the subsequence prediction results are superimposed to obtain the final prediction results. The comparison test with the actual load data of a certain place proves that the model has high prediction accuracy and can provide reliable basis for the stable operation of power system.
In order to solve the problem of insufficient forecasting accuracy caused by many factors such as nonlinear and time series of load data, a short-term load forecasting model based on CEEMDAN and HBA-BiGRU-SelfAttention was proposed. Firstly, the random forest algorithm is used to extract the features of meteorological factors, which ensures the characteristics of data and reduces the complexity of data. Secondly, the adaptive noise complete set empirical mode decomposition algorithm is used to decompose the original load data, and some relatively stable modal components are obtained. Then, meteorological factors and modal components extracted by feature are taken as input data, and BiGRU-SelfAttention model is used for prediction. For the problem that it is difficult to select the optimal solution for hyperparameters of BiGRU-SelfAttention model, The Honey Badger algorithm is introduced to optimize the hyperparameters of BiGRU-SelfAttention model. Finally, the subsequence prediction results are superimposed to obtain the final prediction results. The comparison test with the actual load data of a certain place proves that the model has high prediction accuracy and can provide reliable basis for the stable operation of power system.
Abstract:
To achieve autonomous navigation in unstructured environments, unmanned vehicles need to analyze the traversability of the terrain. Currently, methods based on LiDAR and vision are used for this analysis, but LiDAR systems are limited by sparse point clouds and high costs, and traditional vision approaches fail to effectively capture and express the three-dimensional spatial conditions of the scene. Addressing these challenges, this paper introduces for the first time a method for analyzing traversability in unstructured environments based on Occupancy Prediction, named WildOcc. WildOcc extracts multi-scale features from monocular RGB images, projects 3D occupancy labels onto the images, and introduces a road attention mechanism to query points and fuse information to obtain 3D features, which are then output as traversable areas through a decoder and semantic segmentation head. To accurately estimate the three-dimensional traversability of the environment, WildOcc uses 3D occupancy labels for supervision; due to the sparsity of point cloud data in unstructured environments, this paper designs a data enhancement module called Dense Label Generate (DLG) to produce dense occupancy labels, improving the accuracy of the supervision results. Based on the DLG module, this paper produces the first dataset usable for occupancy prediction in unstructured environments. Comprehensive experiments conducted on this dataset show that, relative to occupancy prediction methods designed for structured environments, the DLG module improves the mIoU by 0.7%, and jointly with WildOcc, enhances the mIoU by 1%, effectively increasing the prediction accuracy and robustness.
To achieve autonomous navigation in unstructured environments, unmanned vehicles need to analyze the traversability of the terrain. Currently, methods based on LiDAR and vision are used for this analysis, but LiDAR systems are limited by sparse point clouds and high costs, and traditional vision approaches fail to effectively capture and express the three-dimensional spatial conditions of the scene. Addressing these challenges, this paper introduces for the first time a method for analyzing traversability in unstructured environments based on Occupancy Prediction, named WildOcc. WildOcc extracts multi-scale features from monocular RGB images, projects 3D occupancy labels onto the images, and introduces a road attention mechanism to query points and fuse information to obtain 3D features, which are then output as traversable areas through a decoder and semantic segmentation head. To accurately estimate the three-dimensional traversability of the environment, WildOcc uses 3D occupancy labels for supervision; due to the sparsity of point cloud data in unstructured environments, this paper designs a data enhancement module called Dense Label Generate (DLG) to produce dense occupancy labels, improving the accuracy of the supervision results. Based on the DLG module, this paper produces the first dataset usable for occupancy prediction in unstructured environments. Comprehensive experiments conducted on this dataset show that, relative to occupancy prediction methods designed for structured environments, the DLG module improves the mIoU by 0.7%, and jointly with WildOcc, enhances the mIoU by 1%, effectively increasing the prediction accuracy and robustness.
Abstract:
In real-world industrial IoT environments, sensor data is often contaminated by external noise, making it difficult to obtain clean data, which affects the accuracy of data-driven time series prediction tasks. To address this issue, this paper proposes a novel time series prediction framework that combines an improved contrast blind denoising autoencoder (CBDMAE) and a TCN-Transformer network, called MoCo-CBDAE-TCN-Transformer. The framework enhances its ability to capture dynamic features of time series data by introducing additional momentum encoders, dynamic queues, and information noise-contrastive estimation (InfoNCE) regularization. It effectively utilizes historical negative sample information by capturing and comparing temporal correlations and noise features. Without prior knowledge of noise and clean sensor data, the network achieves blind denoising of sensor data by capturing and comparing temporal correlations and noise features. The denoised data is then predicted using a TCN-Transformer network, which combines the advantages of TCN"s residual connections and dilated convolutions, as well as Transformer"s attention mechanism, significantly improving prediction accuracy and efficiency. Finally, the experimental results show that, compared with traditional noise removal methods and time series prediction models, the model designed in this paper can achieve better noise removal effects and higher prediction accuracy through simulation verification on a public four-cylinder process dataset. Its real-time processing capability is suitable for deployment in actual industrial environments and provides an effective technical solution for data processing and analysis in industrial Internet of Things.
In real-world industrial IoT environments, sensor data is often contaminated by external noise, making it difficult to obtain clean data, which affects the accuracy of data-driven time series prediction tasks. To address this issue, this paper proposes a novel time series prediction framework that combines an improved contrast blind denoising autoencoder (CBDMAE) and a TCN-Transformer network, called MoCo-CBDAE-TCN-Transformer. The framework enhances its ability to capture dynamic features of time series data by introducing additional momentum encoders, dynamic queues, and information noise-contrastive estimation (InfoNCE) regularization. It effectively utilizes historical negative sample information by capturing and comparing temporal correlations and noise features. Without prior knowledge of noise and clean sensor data, the network achieves blind denoising of sensor data by capturing and comparing temporal correlations and noise features. The denoised data is then predicted using a TCN-Transformer network, which combines the advantages of TCN"s residual connections and dilated convolutions, as well as Transformer"s attention mechanism, significantly improving prediction accuracy and efficiency. Finally, the experimental results show that, compared with traditional noise removal methods and time series prediction models, the model designed in this paper can achieve better noise removal effects and higher prediction accuracy through simulation verification on a public four-cylinder process dataset. Its real-time processing capability is suitable for deployment in actual industrial environments and provides an effective technical solution for data processing and analysis in industrial Internet of Things.
Abstract:
Bi2O2CO3/Bi4Ti3O12 composite heterojunction photocatalyst was prepared using in-situ generation method. The crystal structure, microstructure, elemental composition, and optoelectronic properties of the prepared material were studied through a series of characterization methods. The degradation activity of the prepared photocatalyst was studied by degrading the antibiotic ofloxacin under visible light. The experimental results show that the degradation activity of the composite photocatalyst is superior to that of a single component photocatalyst, and the optimal composite sample BTC-2 has an apparent rate constant that is 6.3 times and 2.2 times that of pure Bi2O2CO3 and Bi4Ti3O12, respectively. The 4 cycles of experiments indicate that the composite sample has good stability. According to the electrochemical test results, it was found that the BOC/BTO composite photocatalyst has good photo generated carrier separation efficiency and excellent charge transfer efficiency, thus exhibiting good degradation activity towards ofloxacin. The free radical capture experiment indicates that the main active species of BOC/BTO composites in photocatalytic experiments are holes and superoxide radicals. In addition, charge transfer pathways and mechanisms for enhancing activity were proposed through free radical capture experiments and band theory.
Bi2O2CO3/Bi4Ti3O12 composite heterojunction photocatalyst was prepared using in-situ generation method. The crystal structure, microstructure, elemental composition, and optoelectronic properties of the prepared material were studied through a series of characterization methods. The degradation activity of the prepared photocatalyst was studied by degrading the antibiotic ofloxacin under visible light. The experimental results show that the degradation activity of the composite photocatalyst is superior to that of a single component photocatalyst, and the optimal composite sample BTC-2 has an apparent rate constant that is 6.3 times and 2.2 times that of pure Bi2O2CO3 and Bi4Ti3O12, respectively. The 4 cycles of experiments indicate that the composite sample has good stability. According to the electrochemical test results, it was found that the BOC/BTO composite photocatalyst has good photo generated carrier separation efficiency and excellent charge transfer efficiency, thus exhibiting good degradation activity towards ofloxacin. The free radical capture experiment indicates that the main active species of BOC/BTO composites in photocatalytic experiments are holes and superoxide radicals. In addition, charge transfer pathways and mechanisms for enhancing activity were proposed through free radical capture experiments and band theory.
Abstract:
In few-shot learning scenarios, Model-Agnostic Meta-Learning (MAML) is renowned for its flexibility and applicability independent of specific base model. However, MAML and its variants lack the ability to adaptively adjust task-specific features, such as feature channels, and rely on global initial parameters with fixed inner-loop step numbers, resulting in a training process that lacks interpretability. This paper proposes a rapid, interpretable, and adaptive method using logistic regression and channel attention (ML-LRCA). The goal is to enhance the adaptability and interpretability of the framework by adaptively adjusting task-specific channel weights and utilizing logistic regression to swiftly adapt to task-specific adaptive loss functions. By conducting experiments on multiple open-source datasets for few-shot classification, cross-domain few-shot classification, and few-shot regression, the results indicate that the proposed ML-LRCA method achieves significant performance improvement.
In few-shot learning scenarios, Model-Agnostic Meta-Learning (MAML) is renowned for its flexibility and applicability independent of specific base model. However, MAML and its variants lack the ability to adaptively adjust task-specific features, such as feature channels, and rely on global initial parameters with fixed inner-loop step numbers, resulting in a training process that lacks interpretability. This paper proposes a rapid, interpretable, and adaptive method using logistic regression and channel attention (ML-LRCA). The goal is to enhance the adaptability and interpretability of the framework by adaptively adjusting task-specific channel weights and utilizing logistic regression to swiftly adapt to task-specific adaptive loss functions. By conducting experiments on multiple open-source datasets for few-shot classification, cross-domain few-shot classification, and few-shot regression, the results indicate that the proposed ML-LRCA method achieves significant performance improvement.
Abstract:
Abstract: When traditional convolutional neural networks are used to classify the spectrum images of impact echo signals, they are faced with the problems of insufficient feature extraction ability, unbalanced data sets and small scale. This paper presents the Multi-scale Hybrid Attention and Spatial Channel Reconstruction Convolutional Neural Network (MHA-SCConvNet). The model initiates with a multi-scale hybrid attention module that extracts spectral features at diverse scales, significantly sharpening the focus on essential waveform information. This is followed by the integration of a Spatial and Channel Reconstruction Convolution (SCConv) module, designed to optimize the representation of image features and reduce redundancy effectively. Additionally, the paper introduces the Gradient and Distribution Harmonized Margin Loss (GDHM Loss), a dynamic hybrid loss function tailored to address the challenges of classifying both hard-to-classify and minority class samples. The MHA-SCConvNet was rigorously evaluated on a proprietary dataset, where it demonstrated a notable accuracy of 94.58%, outperforming established models such as AlexNet, VGGNet, and GoogLeNet. Experimental results validate the MHA-SCConvNet's superior capability in enhancing the accuracy and efficiency of impact echo spectral image classification.
Abstract: When traditional convolutional neural networks are used to classify the spectrum images of impact echo signals, they are faced with the problems of insufficient feature extraction ability, unbalanced data sets and small scale. This paper presents the Multi-scale Hybrid Attention and Spatial Channel Reconstruction Convolutional Neural Network (MHA-SCConvNet). The model initiates with a multi-scale hybrid attention module that extracts spectral features at diverse scales, significantly sharpening the focus on essential waveform information. This is followed by the integration of a Spatial and Channel Reconstruction Convolution (SCConv) module, designed to optimize the representation of image features and reduce redundancy effectively. Additionally, the paper introduces the Gradient and Distribution Harmonized Margin Loss (GDHM Loss), a dynamic hybrid loss function tailored to address the challenges of classifying both hard-to-classify and minority class samples. The MHA-SCConvNet was rigorously evaluated on a proprietary dataset, where it demonstrated a notable accuracy of 94.58%, outperforming established models such as AlexNet, VGGNet, and GoogLeNet. Experimental results validate the MHA-SCConvNet's superior capability in enhancing the accuracy and efficiency of impact echo spectral image classification.
Abstract:
This study investigated the tendency of the human brain to respond to a standardized corpus of speech by measuring changes in speech-related EEG signals. Sixteen subjects listened to 120 standardized speech items for 8 seconds each, with 1-2 second intervals between the items and played randomly. The EEG signals were extracted during the listening process, the EEG signals in the frequency band of 1-40 Hz were preprocessed and analyzed, then compared them with speech signals. The results show that subjects have the same trend of EEG response when they hear the same standard speech. In addition, this paper also measured the phase difference between the EEG and speech signals by the phase-locked value method, which proved the functional connectivity between the EEG signals and speech quality. Furthermore, the EEG signals could distinguish the speech quality with a correct rate of 99.62%.
This study investigated the tendency of the human brain to respond to a standardized corpus of speech by measuring changes in speech-related EEG signals. Sixteen subjects listened to 120 standardized speech items for 8 seconds each, with 1-2 second intervals between the items and played randomly. The EEG signals were extracted during the listening process, the EEG signals in the frequency band of 1-40 Hz were preprocessed and analyzed, then compared them with speech signals. The results show that subjects have the same trend of EEG response when they hear the same standard speech. In addition, this paper also measured the phase difference between the EEG and speech signals by the phase-locked value method, which proved the functional connectivity between the EEG signals and speech quality. Furthermore, the EEG signals could distinguish the speech quality with a correct rate of 99.62%.
Abstract:
Aiming at the problem that in indoor dynamic environments, traditional visual SLAM algorithms are affected by a large amount of meaningless information, which leads to a decrease in localization accuracy and poor robustness, a visual SLAM algorithm PLYO-SLAM for indoor dynamic scenes based on YOLOv8 is proposed. In this paper, the algorithm introduces EDLines line segment detection algorithm in the tracking thread of ORB-SLAM3 algorithm, and adds a new Dynamic region detection thread. The dynamic region detection thread consists of a YOLOv8n-seg instance segmentation network, where the instance segmentation empowers the dynamic scene semantic information and generates dynamic masks, and at the same time eliminates the point and line features of dynamic regions, and utilizes the geometric constraints to further filter the missing dynamic feature points outside the segmentation masks. Experimental validation using the publicly available dataset TUM shows that compared to the ORB-SLAM3 algorithm, the PLYO-SLAM algorithm improves the localization accuracy in dynamic environments by an average of 75.99% and reaches a maximum of 96.75%.
Aiming at the problem that in indoor dynamic environments, traditional visual SLAM algorithms are affected by a large amount of meaningless information, which leads to a decrease in localization accuracy and poor robustness, a visual SLAM algorithm PLYO-SLAM for indoor dynamic scenes based on YOLOv8 is proposed. In this paper, the algorithm introduces EDLines line segment detection algorithm in the tracking thread of ORB-SLAM3 algorithm, and adds a new Dynamic region detection thread. The dynamic region detection thread consists of a YOLOv8n-seg instance segmentation network, where the instance segmentation empowers the dynamic scene semantic information and generates dynamic masks, and at the same time eliminates the point and line features of dynamic regions, and utilizes the geometric constraints to further filter the missing dynamic feature points outside the segmentation masks. Experimental validation using the publicly available dataset TUM shows that compared to the ORB-SLAM3 algorithm, the PLYO-SLAM algorithm improves the localization accuracy in dynamic environments by an average of 75.99% and reaches a maximum of 96.75%.
Shao Pengfei, Hu Kai, Liu Ziran, Feng Xinyan, Ma Keyu, Zhang Qi, Jiang Shanshan, Xia Min, Ye Xiaoling
Abstract:
The detection of greenhouse gas plumes is a key task of carbon emission monitoring. In the current relevant algorithms, the segmentation accuracy of plume edge details still has room for improvement. In this paper, a multi-scale attention mechanism based plume detection network (MAPDNet) is proposed. Firstly, a multi-scale pooled strip Convolution module (MPSCM) is proposed to compensate the high-level semantic information diluted in the downsampling operation of the encoder, enhance the main plume profile information, and reduce the loss of detail information. Then, the relational attention module (RAM) establishes the global and local information interaction between features, enhances the useful features, reduces the proportion of redundant information in the feature map, and improves the segmentation accuracy of the model. In addition, a context feature fusion module (CFFM) is proposed in this paper so that the network can understand the context feature information better, align the features better, and recover the spatial position information while recovering the image resolution. Finally, the edge enhancement module (BEM) implements end-to-end training, enhances the details of the output segmentation boundary, and further improves the segmentation accuracy. The experimental results show that MAPDNet model can detect carbon plumes from XCO2 plume images, and has good segmentation performance on simulated plume data sets, and its segmentation accuracy is better than the existing methods.
The detection of greenhouse gas plumes is a key task of carbon emission monitoring. In the current relevant algorithms, the segmentation accuracy of plume edge details still has room for improvement. In this paper, a multi-scale attention mechanism based plume detection network (MAPDNet) is proposed. Firstly, a multi-scale pooled strip Convolution module (MPSCM) is proposed to compensate the high-level semantic information diluted in the downsampling operation of the encoder, enhance the main plume profile information, and reduce the loss of detail information. Then, the relational attention module (RAM) establishes the global and local information interaction between features, enhances the useful features, reduces the proportion of redundant information in the feature map, and improves the segmentation accuracy of the model. In addition, a context feature fusion module (CFFM) is proposed in this paper so that the network can understand the context feature information better, align the features better, and recover the spatial position information while recovering the image resolution. Finally, the edge enhancement module (BEM) implements end-to-end training, enhances the details of the output segmentation boundary, and further improves the segmentation accuracy. The experimental results show that MAPDNet model can detect carbon plumes from XCO2 plume images, and has good segmentation performance on simulated plume data sets, and its segmentation accuracy is better than the existing methods.
Abstract:
Under the leader-follower structure, based on the intermittent event-triggered mechanism, this paper investigates the finite-time formation system problem for unmanned surface vessel clusters with second-order dynamics. Firstly, by using adaptive neural network techniques, unknown nonlinear items are estimated in the systems model. Secondly, a kind of formation control protocol is proposed based on segmented intermittent event-triggered and adaptive methods to reduce communication energy consumption and accelerate cluster convergence. Then, a sufficient condition for unmanned surface vessel clusters to rapidly form formations in finite-time is obtained using theories such as Lyapunov stabilization and finite time. Finally, simulation examples validate the superiority and effectiveness of the proposed algorithms in the paper.
Under the leader-follower structure, based on the intermittent event-triggered mechanism, this paper investigates the finite-time formation system problem for unmanned surface vessel clusters with second-order dynamics. Firstly, by using adaptive neural network techniques, unknown nonlinear items are estimated in the systems model. Secondly, a kind of formation control protocol is proposed based on segmented intermittent event-triggered and adaptive methods to reduce communication energy consumption and accelerate cluster convergence. Then, a sufficient condition for unmanned surface vessel clusters to rapidly form formations in finite-time is obtained using theories such as Lyapunov stabilization and finite time. Finally, simulation examples validate the superiority and effectiveness of the proposed algorithms in the paper.
Abstract:
The paper addresses the issue of poor path-tracking control accuracy in autonomous vehicles under special driving conditions (such as icy and snowy roads, rainy surfaces, and high-speed lane changes). A Linear Quadratic Regulator (LQR) controller based on the Whale Optimization Algorithm (WOA), referred to as the WOA-LQR controller, is designed to tackle this problem. Initially, a tracking error model is established based on a two-degree-of-freedom vehicle dynamics model, and a discrete LQR controller is designed on this basis. Feedforward control is employed to eliminate errors caused by system simplification. To address the problem of low tracking accuracy and vehicle instability under special driving conditions due to the poor adaptability of the LQR controller with fixed weight coefficients, an adaptive weight adjustment strategy for the LQR controller is proposed based on WOA. This strategy takes into account the influence of lateral acceleration and front wheel steering angle on vehicle stability and sets corresponding weight coefficients for the evaluation indices, which include lateral error and yaw angle error. A fitness function with the minimum target value is designed to optimize the LQR weight coefficients using WOA. Finally, the WOA-LQR controller is validated through path-tracking simulations under various conditions using Carsim/Simulink co-simulation. The results demonstrate that the proposed control strategy provides excellent tracking performance under complex driving conditions, significantly improves the control accuracy during path tracking, and exhibits strong robustness.
The paper addresses the issue of poor path-tracking control accuracy in autonomous vehicles under special driving conditions (such as icy and snowy roads, rainy surfaces, and high-speed lane changes). A Linear Quadratic Regulator (LQR) controller based on the Whale Optimization Algorithm (WOA), referred to as the WOA-LQR controller, is designed to tackle this problem. Initially, a tracking error model is established based on a two-degree-of-freedom vehicle dynamics model, and a discrete LQR controller is designed on this basis. Feedforward control is employed to eliminate errors caused by system simplification. To address the problem of low tracking accuracy and vehicle instability under special driving conditions due to the poor adaptability of the LQR controller with fixed weight coefficients, an adaptive weight adjustment strategy for the LQR controller is proposed based on WOA. This strategy takes into account the influence of lateral acceleration and front wheel steering angle on vehicle stability and sets corresponding weight coefficients for the evaluation indices, which include lateral error and yaw angle error. A fitness function with the minimum target value is designed to optimize the LQR weight coefficients using WOA. Finally, the WOA-LQR controller is validated through path-tracking simulations under various conditions using Carsim/Simulink co-simulation. The results demonstrate that the proposed control strategy provides excellent tracking performance under complex driving conditions, significantly improves the control accuracy during path tracking, and exhibits strong robustness.
Abstract:
Accurately detecting tea buds is the key to achieving automation and intelligence in tea bud harvesting. However, to accurately identify tea bud images, it is necessary to overcome the problem of tea bud colors being similar to the background and the target size being too small. Therefore, this article studies a YOLOv5s model based on hybrid attention mechanism and applies it to tea bud detection. This article makes optimizations in the following two aspects: firstly, a hybrid attention mechanism (HAM) is proposed and added to the YOLOv5s backbone network, which enables the network to focus on the target area, extract features more fully, and improve the accuracy of object recognition by the model. Secondly, by introducing normalized Wasserstein distance (NWD) as a new metric and combining it with the existing CIoU loss function. The NWD loss function calculates the similarity between the bounding boxes based on their corresponding Gaussian distributions, thereby improving the model's accuracy in detecting small targets in images. The experimental results show that compared with the original YOLOv5s model, the improved model mAP0.5 increased by 0.9%, mAP0.5:0.95 increased by 1.3%, while the number of parameters only increased by 0.044×106. These results confirm the effectiveness of the proposed method in achieving precise tea picking recognition, providing technical reference for intelligent tea picking in practical scenarios.
Accurately detecting tea buds is the key to achieving automation and intelligence in tea bud harvesting. However, to accurately identify tea bud images, it is necessary to overcome the problem of tea bud colors being similar to the background and the target size being too small. Therefore, this article studies a YOLOv5s model based on hybrid attention mechanism and applies it to tea bud detection. This article makes optimizations in the following two aspects: firstly, a hybrid attention mechanism (HAM) is proposed and added to the YOLOv5s backbone network, which enables the network to focus on the target area, extract features more fully, and improve the accuracy of object recognition by the model. Secondly, by introducing normalized Wasserstein distance (NWD) as a new metric and combining it with the existing CIoU loss function. The NWD loss function calculates the similarity between the bounding boxes based on their corresponding Gaussian distributions, thereby improving the model's accuracy in detecting small targets in images. The experimental results show that compared with the original YOLOv5s model, the improved model mAP0.5 increased by 0.9%, mAP0.5:0.95 increased by 1.3%, while the number of parameters only increased by 0.044×106. These results confirm the effectiveness of the proposed method in achieving precise tea picking recognition, providing technical reference for intelligent tea picking in practical scenarios.
Abstract:
The Automatic guided vehicle needs to avoid not only static obstacles, but also moving obstacles in the process of driving, and the commonly used dynamic window local path planning algorithm is difficult to meet the actual needs of the success rate of obstacle avoidance and obstacle avoidance efficiency in such cases. In this paper, a new adaptive dynamic window local path planning algorithm is proposed based on the analytic hierarchy process. Firstly, the analytic hierarchy process is used to calculate the value of the weight factor in the dynamic window approach. Then, a collision safety index is proposed to evaluate the safety of the current running state, and the self-feedback mechanism between the weight factor and the environmental factors is established by combining the safety membership function of the relevant parameters. Finally, the obstacle avoidance route is planned based on the created adaptive dynamic window approach. The simulation results show that the proposed adaptive dynamic window approach can adaptively adjust the weight factor of the evaluation function, effectively avoid static obstacles and moving obstacles, and has been significantly optimized in terms of travel time, path length and safety.
The Automatic guided vehicle needs to avoid not only static obstacles, but also moving obstacles in the process of driving, and the commonly used dynamic window local path planning algorithm is difficult to meet the actual needs of the success rate of obstacle avoidance and obstacle avoidance efficiency in such cases. In this paper, a new adaptive dynamic window local path planning algorithm is proposed based on the analytic hierarchy process. Firstly, the analytic hierarchy process is used to calculate the value of the weight factor in the dynamic window approach. Then, a collision safety index is proposed to evaluate the safety of the current running state, and the self-feedback mechanism between the weight factor and the environmental factors is established by combining the safety membership function of the relevant parameters. Finally, the obstacle avoidance route is planned based on the created adaptive dynamic window approach. The simulation results show that the proposed adaptive dynamic window approach can adaptively adjust the weight factor of the evaluation function, effectively avoid static obstacles and moving obstacles, and has been significantly optimized in terms of travel time, path length and safety.
Abstract:
The increasing load density and the higher proportion of external power feed-in continuously raise the level of short-circuit currents in the receiving-end transmission network, affecting the safe operation of the receiving-end grid. To address this issue, this paper proposes an optimization method for the receiving-end grid with multi-DC feed-in based on complex network theory. We investigate and propose an improved community detection algorithm based on complex network theory to search for critical interconnection lines in the receiving-end grid, forming a pre-disconnection line set. Targeting the improvement of short-circuit current level, the ratio of multiple feed-in short-circuit, and the active loss of the grid, we establish a multi-objective optimization model for the receiving-end grid framework. To enhance the accuracy of the model, we consider the impact of uncertainty in renewable energy output in the grid operation mode, utilizing non-parametric kernel density estimation and Frank-Copula function to generate scenarios of wind and solar output and cluster to obtain a set of typical scenarios. The NSGA-II algorithm is employed to solve the model, obtaining a Pareto optimal solution set, from which the optimal decision solution satisfying N-1 verification is selected. Finally, the effectiveness of the proposed method is validated through case studies.
The increasing load density and the higher proportion of external power feed-in continuously raise the level of short-circuit currents in the receiving-end transmission network, affecting the safe operation of the receiving-end grid. To address this issue, this paper proposes an optimization method for the receiving-end grid with multi-DC feed-in based on complex network theory. We investigate and propose an improved community detection algorithm based on complex network theory to search for critical interconnection lines in the receiving-end grid, forming a pre-disconnection line set. Targeting the improvement of short-circuit current level, the ratio of multiple feed-in short-circuit, and the active loss of the grid, we establish a multi-objective optimization model for the receiving-end grid framework. To enhance the accuracy of the model, we consider the impact of uncertainty in renewable energy output in the grid operation mode, utilizing non-parametric kernel density estimation and Frank-Copula function to generate scenarios of wind and solar output and cluster to obtain a set of typical scenarios. The NSGA-II algorithm is employed to solve the model, obtaining a Pareto optimal solution set, from which the optimal decision solution satisfying N-1 verification is selected. Finally, the effectiveness of the proposed method is validated through case studies.
Abstract:
With the large-scale integration of renewable energy sources and the continuous growth in electricity demand, the operating environment of power grids has become increasingly complex and variable. This change has introduced numerous uncertainties, making it difficult for traditional model-based or purely data-driven grid scheduling methods to formulate accurate and reliable scheduling plans. To address this challenge, this paper proposes an adaptive cruising method for grid scheduling that integrates data and knowledge. Firstly, a mathematical model for the adaptive cruising problem in power system dispatch is constructed, and a deep reinforcement learning training framework is built based on this model. Furthermore, this article proposes an improved proximal policy optimization algorithm for offline training of dispatch agent. By integrating professional knowledge related to scheduling regulations, the proposed algorithm can guide agents to learn towards optimization. Finally, an in-depth analysis of simulation examples validates the effectiveness and superiority of the proposed method.
With the large-scale integration of renewable energy sources and the continuous growth in electricity demand, the operating environment of power grids has become increasingly complex and variable. This change has introduced numerous uncertainties, making it difficult for traditional model-based or purely data-driven grid scheduling methods to formulate accurate and reliable scheduling plans. To address this challenge, this paper proposes an adaptive cruising method for grid scheduling that integrates data and knowledge. Firstly, a mathematical model for the adaptive cruising problem in power system dispatch is constructed, and a deep reinforcement learning training framework is built based on this model. Furthermore, this article proposes an improved proximal policy optimization algorithm for offline training of dispatch agent. By integrating professional knowledge related to scheduling regulations, the proposed algorithm can guide agents to learn towards optimization. Finally, an in-depth analysis of simulation examples validates the effectiveness and superiority of the proposed method.
Abstract:
针对缺陷语义分割任务中缺陷与背景差异小以及同类缺陷间差异大的问题，本文提出了一种基于轻量化卷积块注意力Transformer的表面缺陷检测算法，即LCBAFormer，该算法旨在提高不同类型缺陷的分割准确率。本文首先设计了轻量化卷积块注意力模块LCBAM，通过结合通道注意力模块和空间注意力模块，提取有效的通道信息和空间信息，使模型更专注于局部缺陷特征信息，增强不同缺陷间的特征差异，减少同类缺陷间的差异。其次，本文提出轻量化的语义注入模块SIM，以逐步融合多尺度特征信息，提升网络对不同缺陷的定位和区分能力。实验结果表明，在NEU-Seg钢带缺陷数据集和MT-Defect磁瓦缺陷数据集上，本文提出的算法平均交并比（mIoU）分别为84.75%和79.46%mIoU，平均召回率（mRec）分别为92.29%和87.50%，平均F1分数（mF1）分别为91.52%和88.08%mF1，此外，算法的计算复杂度较低，每秒十亿次浮点运算次数（GFLOPs）分别为1.03和2.65。与主流算法相比，本文方法在参数量和检测性能上做到了较好的平衡，具有更小参数量和更好的分割结果。
针对缺陷语义分割任务中缺陷与背景差异小以及同类缺陷间差异大的问题，本文提出了一种基于轻量化卷积块注意力Transformer的表面缺陷检测算法，即LCBAFormer，该算法旨在提高不同类型缺陷的分割准确率。本文首先设计了轻量化卷积块注意力模块LCBAM，通过结合通道注意力模块和空间注意力模块，提取有效的通道信息和空间信息，使模型更专注于局部缺陷特征信息，增强不同缺陷间的特征差异，减少同类缺陷间的差异。其次，本文提出轻量化的语义注入模块SIM，以逐步融合多尺度特征信息，提升网络对不同缺陷的定位和区分能力。实验结果表明，在NEU-Seg钢带缺陷数据集和MT-Defect磁瓦缺陷数据集上，本文提出的算法平均交并比（mIoU）分别为84.75%和79.46%mIoU，平均召回率（mRec）分别为92.29%和87.50%，平均F1分数（mF1）分别为91.52%和88.08%mF1，此外，算法的计算复杂度较低，每秒十亿次浮点运算次数（GFLOPs）分别为1.03和2.65。与主流算法相比，本文方法在参数量和检测性能上做到了较好的平衡，具有更小参数量和更好的分割结果。
Abstract:
Considering the spectral characteristics of substation noise, an enhanced generative fixed filter active noise control (Enhanced Generative Fixed Filter Active Noise Control, EGFANC) method was introduced to improve the limitations of adaptive algorithms such as slow convergence speed, weak tracking ability, and large computational complexity. A lightweight one-dimensional convolutional neural network (1D CNN) was used to output the weight vector based on noise frame information, then the weight vector was combined with sub control filters to adaptively generate suitable control filters for various types of noise. The simulation results show that the EGFANC method has better noise reduction performance and robustness in dealing with dynamic noise and transformer harmonic noise. In addition, EGFANC can significantly reduce convergence time by selecting appropriate pre trained control filters for different types of noise.
Considering the spectral characteristics of substation noise, an enhanced generative fixed filter active noise control (Enhanced Generative Fixed Filter Active Noise Control, EGFANC) method was introduced to improve the limitations of adaptive algorithms such as slow convergence speed, weak tracking ability, and large computational complexity. A lightweight one-dimensional convolutional neural network (1D CNN) was used to output the weight vector based on noise frame information, then the weight vector was combined with sub control filters to adaptively generate suitable control filters for various types of noise. The simulation results show that the EGFANC method has better noise reduction performance and robustness in dealing with dynamic noise and transformer harmonic noise. In addition, EGFANC can significantly reduce convergence time by selecting appropriate pre trained control filters for different types of noise.
Abstract:
Aiming at the difficulties such as poor formation stability and low passage efficiency of smart platooning (SP) when passing through unsignalized intersections, an intelligent vehicle platooning control method based on Virtual Leaders (VLs) has been proposed. Firstly, taking the SP as the research object, improving the traditional artificial potential function (IAPF),using nonlinear functional relationships to describe the vehicle-vehicle forces within the SP system. Secondly, based on the idea of virtual navigation, an improved artificial potential function flocking control algorithm based on virtual leaders (IAPFFCA-VLs) for SP forming has been designed, and the stability of the algorithm has been proved by Lyapunov stability method and LaSalle invariance principle. Finally, using Matlab to simulate the SP forming control and compared with the flocking control algorithm incorporating IAPF (FCA-IAPF). The simulation results show that the FCA-IAPF can realize vehicle-vehicle distance control and collision avoidance, but with the increase of the number of intelligent vehicles, there is the problem of local optimal ‘chunking’ and can not form a regular SP. Compared to FCA-IAPF, the overall performance of the IAPFFCA-VLs algorithm are optimized, which can realize different initial states, different numbers of vehicles with the same destination to maintain reasonable spacing, avoiding collision while speed gradual to reach consistency, the positional distance deviation end-value between all intelligent vehicles and their respective VL is less than 0.1m, the longitudinal velocity deviation end-value is less than 1m/s, and the transverse velocity deviation end-value tends to be 0.001m/s, safe, rapid and stable constrained as a regular SP and drive out of the unsignalized intersection, which is important to improve the unsignalized intersection SP driving stability as well as traffic efficiency.
Aiming at the difficulties such as poor formation stability and low passage efficiency of smart platooning (SP) when passing through unsignalized intersections, an intelligent vehicle platooning control method based on Virtual Leaders (VLs) has been proposed. Firstly, taking the SP as the research object, improving the traditional artificial potential function (IAPF),using nonlinear functional relationships to describe the vehicle-vehicle forces within the SP system. Secondly, based on the idea of virtual navigation, an improved artificial potential function flocking control algorithm based on virtual leaders (IAPFFCA-VLs) for SP forming has been designed, and the stability of the algorithm has been proved by Lyapunov stability method and LaSalle invariance principle. Finally, using Matlab to simulate the SP forming control and compared with the flocking control algorithm incorporating IAPF (FCA-IAPF). The simulation results show that the FCA-IAPF can realize vehicle-vehicle distance control and collision avoidance, but with the increase of the number of intelligent vehicles, there is the problem of local optimal ‘chunking’ and can not form a regular SP. Compared to FCA-IAPF, the overall performance of the IAPFFCA-VLs algorithm are optimized, which can realize different initial states, different numbers of vehicles with the same destination to maintain reasonable spacing, avoiding collision while speed gradual to reach consistency, the positional distance deviation end-value between all intelligent vehicles and their respective VL is less than 0.1m, the longitudinal velocity deviation end-value is less than 1m/s, and the transverse velocity deviation end-value tends to be 0.001m/s, safe, rapid and stable constrained as a regular SP and drive out of the unsignalized intersection, which is important to improve the unsignalized intersection SP driving stability as well as traffic efficiency.
Abstract:
The mathematical model of flexible job-shop scheduling problem is established to minimize the maximum completion time of all workpieces under the constraints of processing sequence and processing time. A three population co-evolution algorithm was proposed to solve the model. Based on the three-way tournament method, the population was divided into superior, medium and inferior subpopulations, and the corresponding search strategy was designed according to the individual characteristics of different subpopulations. The superior subpopulation uses load balance and variable neighborhood descent local search of critical path to improve the solution accuracy and excavate better solutions. In the medium subpopulation, adaptive Jaya operation was used to avoid the bad in the early stage of evolution, and the maintenance of population diversity was emphasized in the middle and late stages. Multiple cross global search was adopted for inferior subpopulation, and the crossover operators that could generate viable individuals were designed for different gene strings. At the same time, individuals in other subpopulations were taken as crossover objects to strengthen the cooperative interaction among subpopulations. A large number of experimental results in standard test examples and production instances show that the proposed algorithm has significantly better performance than the representative algorithm in most cases.
The mathematical model of flexible job-shop scheduling problem is established to minimize the maximum completion time of all workpieces under the constraints of processing sequence and processing time. A three population co-evolution algorithm was proposed to solve the model. Based on the three-way tournament method, the population was divided into superior, medium and inferior subpopulations, and the corresponding search strategy was designed according to the individual characteristics of different subpopulations. The superior subpopulation uses load balance and variable neighborhood descent local search of critical path to improve the solution accuracy and excavate better solutions. In the medium subpopulation, adaptive Jaya operation was used to avoid the bad in the early stage of evolution, and the maintenance of population diversity was emphasized in the middle and late stages. Multiple cross global search was adopted for inferior subpopulation, and the crossover operators that could generate viable individuals were designed for different gene strings. At the same time, individuals in other subpopulations were taken as crossover objects to strengthen the cooperative interaction among subpopulations. A large number of experimental results in standard test examples and production instances show that the proposed algorithm has significantly better performance than the representative algorithm in most cases.
Abstract:
To address the issue that road crack detection algorithms often adopt a two-stage approach, which first obtains the bounding box of the crack area from the image and then further segments the crack mask from the detection box using semantic segmentation techniques, note that these two processes are independent of each other and exhibit low efficiency in actual production. This paper proposes an end-to-end integrated road crack detection method and adopts a more lightweight crack backbone feature extraction network. To tackle the problem of limited generalization ability of models in practical scenarios due to the relatively simple scenes in existing road crack public datasets, a fusion method combining a progressive pyramid and spatial adaptive module for crack feature fusion is proposed. This method aims to enhance the detection ability of small target cracks in complex scenes. The proposed model was trained on a collected dataset of urban complex street scenes, achieving a testing accuracy of 86.3% and a recall rate of 84.1%, thereby demonstrating the feasibility of the proposed method in practical applications.
To address the issue that road crack detection algorithms often adopt a two-stage approach, which first obtains the bounding box of the crack area from the image and then further segments the crack mask from the detection box using semantic segmentation techniques, note that these two processes are independent of each other and exhibit low efficiency in actual production. This paper proposes an end-to-end integrated road crack detection method and adopts a more lightweight crack backbone feature extraction network. To tackle the problem of limited generalization ability of models in practical scenarios due to the relatively simple scenes in existing road crack public datasets, a fusion method combining a progressive pyramid and spatial adaptive module for crack feature fusion is proposed. This method aims to enhance the detection ability of small target cracks in complex scenes. The proposed model was trained on a collected dataset of urban complex street scenes, achieving a testing accuracy of 86.3% and a recall rate of 84.1%, thereby demonstrating the feasibility of the proposed method in practical applications.
Abstract:
Due to the uncertain characteristics of wind power generation, wind power stations are faced with the challenge of energy balance and output fluctuation when they are online. Aiming at the differential characteristics of wind power and error distribution in different regions, a wind storage day bidding optimization model based on kernel density and Copula function is proposed. Firstly, the kernel density estimation method is used to calculate the probability density function of the scenic power, and the Archimedes cluster Copula function is introduced to solve the joint distribution function of the scenic power. Then Monte Carlo sampling and K-means clustering method are used to generate the typical output scene. Finally, a pre-day bidding optimization model of wind power station equipped with energy storage considering energy storage peak-valley arbitrage is established. The results show that the proposed model improves the accuracy of describing the output characteristics of the wind-solar power station, realizes a better Internet access and energy storage strategy, verifies the effectiveness of increasing revenue and improving accuracy, and enables the wind-solar power plant cluster to better cope with the output fluctuation problem and improve revenue.
Due to the uncertain characteristics of wind power generation, wind power stations are faced with the challenge of energy balance and output fluctuation when they are online. Aiming at the differential characteristics of wind power and error distribution in different regions, a wind storage day bidding optimization model based on kernel density and Copula function is proposed. Firstly, the kernel density estimation method is used to calculate the probability density function of the scenic power, and the Archimedes cluster Copula function is introduced to solve the joint distribution function of the scenic power. Then Monte Carlo sampling and K-means clustering method are used to generate the typical output scene. Finally, a pre-day bidding optimization model of wind power station equipped with energy storage considering energy storage peak-valley arbitrage is established. The results show that the proposed model improves the accuracy of describing the output characteristics of the wind-solar power station, realizes a better Internet access and energy storage strategy, verifies the effectiveness of increasing revenue and improving accuracy, and enables the wind-solar power plant cluster to better cope with the output fluctuation problem and improve revenue.
Abstract:
The present network for extracting biological echoes from weather radar includes limitations such as a large number of parameters, low efficiency, and the requirement for a large number of samples. This paper offers an improved biological echo extraction network called MFF-PSPnet (Multi-Feature Fusion Pyramid Scene Parsing Network). This network introduces the attention mechanism module and edge extraction module based on the MobilenetV2 backbone, with all features fused for recognition. This modification increases the network's segmentation capabilities in two ways: biological echo main body extraction and biological echo edge detail characterisation. MFF-PSPnet is a lightweight network that uses 10.2% fewer parameters than the conventional biological echo extraction network. Compared to previous models, this network requires less sample capacity and is better suited to small sample situations. On the historical data set, MFF-PSPnet extracted biological echoes with an accuracy of 98.1% and an IoU of 94.8%. The experimental results show that in the historical data of the new generation of Chinese weather radar, the improved model in this work can efficiently extract biological echoes, adapt better to the limited sample environment, and be used to mobile devices. This finding gives useful data and technological support for the future application of weather radar for biological behavior study and protection.
The present network for extracting biological echoes from weather radar includes limitations such as a large number of parameters, low efficiency, and the requirement for a large number of samples. This paper offers an improved biological echo extraction network called MFF-PSPnet (Multi-Feature Fusion Pyramid Scene Parsing Network). This network introduces the attention mechanism module and edge extraction module based on the MobilenetV2 backbone, with all features fused for recognition. This modification increases the network's segmentation capabilities in two ways: biological echo main body extraction and biological echo edge detail characterisation. MFF-PSPnet is a lightweight network that uses 10.2% fewer parameters than the conventional biological echo extraction network. Compared to previous models, this network requires less sample capacity and is better suited to small sample situations. On the historical data set, MFF-PSPnet extracted biological echoes with an accuracy of 98.1% and an IoU of 94.8%. The experimental results show that in the historical data of the new generation of Chinese weather radar, the improved model in this work can efficiently extract biological echoes, adapt better to the limited sample environment, and be used to mobile devices. This finding gives useful data and technological support for the future application of weather radar for biological behavior study and protection.
Abstract:
In safety helmet wear detection, there are problems such as dense targets and occlusion, and existing detection methods perform poorly in terms of accuracy and real-time performance. Existing detection methods often have limitations in terms of both accuracy and real-time performance. To tackle these problems, a lightweight detection model named CA-YOLO is proposed, which is designed to improve the accuracy and real-time performance of detection. Firstly, the backbone network of YOLOv8 is improved using MobileNetv3 network to reduce the number of parameters and computation to improve the detection speed of the network. The DCNv3 module is introduced in the Neck part to improve the extraction efficiency of the model on spatial features. Secondly, a multi-scale feature extraction module and the coordinate attention mechanism module are added to the YOLOv8 network to effectively add global information to enrich feature information to improve the network feature extraction effect. Finally, the CIOU loss is replaced with the Alpha-IOU function, which accelerates the learning process of the target by setting the weight coefficients to further improve the detection accuracy. Experimental results show that compared to the YOLOv8 model and existing classic and novel algorithms, the CA-YOLO model achieves an average detection accuracy of 91.33%, an improvement of 0.54% over the YOLOv8 model. Additionally, The model size and the number of parameters are reduced by 41% and 39%, and the detection speed is increased by 16.9%. Compared to other models, the CA-YOLO model achieves a good balance between accuracy and real-time performance, satisfying the requirements for detecting coal miners" safety helmet wearing.
In safety helmet wear detection, there are problems such as dense targets and occlusion, and existing detection methods perform poorly in terms of accuracy and real-time performance. Existing detection methods often have limitations in terms of both accuracy and real-time performance. To tackle these problems, a lightweight detection model named CA-YOLO is proposed, which is designed to improve the accuracy and real-time performance of detection. Firstly, the backbone network of YOLOv8 is improved using MobileNetv3 network to reduce the number of parameters and computation to improve the detection speed of the network. The DCNv3 module is introduced in the Neck part to improve the extraction efficiency of the model on spatial features. Secondly, a multi-scale feature extraction module and the coordinate attention mechanism module are added to the YOLOv8 network to effectively add global information to enrich feature information to improve the network feature extraction effect. Finally, the CIOU loss is replaced with the Alpha-IOU function, which accelerates the learning process of the target by setting the weight coefficients to further improve the detection accuracy. Experimental results show that compared to the YOLOv8 model and existing classic and novel algorithms, the CA-YOLO model achieves an average detection accuracy of 91.33%, an improvement of 0.54% over the YOLOv8 model. Additionally, The model size and the number of parameters are reduced by 41% and 39%, and the detection speed is increased by 16.9%. Compared to other models, the CA-YOLO model achieves a good balance between accuracy and real-time performance, satisfying the requirements for detecting coal miners" safety helmet wearing.
Abstract:
Aiming at the problem that the path planned by the hybrid A* algorithm is not smooth enough and low efficiency, the evaluation function of the hybrid A* algorithm is optimized, the method of Angle penalty coefficient is introduced to make the path smoother, and the method of node expansion with optimal step length is introduced to improve the search efficiency. Firstly, A hybrid A* algorithm is combined with TEB (Timed Elastic Band) algorithm to generate a hybrid path planning algorithm. Then, the path planning process of Ackermann's steering Robot is simulated based on ROS (Robot Operating System).The results show that the average distance of path planning using mixed A* algorithm and DWA algorithm is 23.83% less than that using A* algorithm and DWA algorithm. The average path planning distance using the mixed A* algorithm and TEB algorithm is 22.49% less than that using the A* algorithm and TEB algorithm. The average path planning time using A* algorithm and TEB algorithm is 6.99% less than that using A* algorithm and DWA algorithm. The average path planning time using the mixed A* algorithm and TEB algorithm is 6.25% less than that using the mixed A* algorithm and DWA algorithm.The path planning efficiency of hybrid A* algorithm combined with TEB algorithm is higher and the planned path is shorter, which proves the effectiveness and superiority of hybrid path planning algorithm. Finally, the path planning experiment of the Ackerman steering robot is carried out in the storage environment. Aiming at the large error between the actual trajectory and the planned trajectory of the robot, a method is proposed to estimate the pose of the obstacle by Kalman filter and track the trajectory of the robot. The results show that the tracking error of Kalman filter is less than 0.2 m.
Aiming at the problem that the path planned by the hybrid A* algorithm is not smooth enough and low efficiency, the evaluation function of the hybrid A* algorithm is optimized, the method of Angle penalty coefficient is introduced to make the path smoother, and the method of node expansion with optimal step length is introduced to improve the search efficiency. Firstly, A hybrid A* algorithm is combined with TEB (Timed Elastic Band) algorithm to generate a hybrid path planning algorithm. Then, the path planning process of Ackermann's steering Robot is simulated based on ROS (Robot Operating System).The results show that the average distance of path planning using mixed A* algorithm and DWA algorithm is 23.83% less than that using A* algorithm and DWA algorithm. The average path planning distance using the mixed A* algorithm and TEB algorithm is 22.49% less than that using the A* algorithm and TEB algorithm. The average path planning time using A* algorithm and TEB algorithm is 6.99% less than that using A* algorithm and DWA algorithm. The average path planning time using the mixed A* algorithm and TEB algorithm is 6.25% less than that using the mixed A* algorithm and DWA algorithm.The path planning efficiency of hybrid A* algorithm combined with TEB algorithm is higher and the planned path is shorter, which proves the effectiveness and superiority of hybrid path planning algorithm. Finally, the path planning experiment of the Ackerman steering robot is carried out in the storage environment. Aiming at the large error between the actual trajectory and the planned trajectory of the robot, a method is proposed to estimate the pose of the obstacle by Kalman filter and track the trajectory of the robot. The results show that the tracking error of Kalman filter is less than 0.2 m.
Abstract:
The robust feature extraction capability of deep learning (DL) has rendered it a popular choice for fault diagnosis in various industries. However, the accuracy of fault diagnosis models is often compromised due to an imbalance between normal and faulty data during the production process. To address this issue, this study proposes an unbalanced bearing fault sample diagnostic method based on a cycle generative adversarial network, using vibration fault diagnosis of rotating machinery rolling bearings as an illustrative example for validation purposes. The method utilizes continuous wavelet transform (CWT) to convert the original vibration signal into time-frequency maps, which are then utilized as input for the cycle generative adversarial network. To address issues related to unstable training and timely convergence of the model, spectral normalization and weight decay techniques are introduced, thereby enhancing the performance of the improved cycle generative adversarial network in generating additional fault samples. Finally, the Swin Transformer model is employed for fault diagnosis and compared with other methods such as Random Forest (RF), Sparse Autoencoder (SAE), Support Vector Machine (SVM), and Convolutional Neural Network (CNN). The diagnostic experiment encompasses sample expansion, imbalanced proportion between normal and faulty samples, as well as scenarios involving limited sample sizes. Experimental results demonstrate that the proposed method can generate higher quality synthetic samples when training data is scarce; moreover, utilizing the Swin Transformer for diagnosis yields superior accuracy compared to alternative approaches such as RF, SAE, SVM, or CNN models – thus showcasing significant potential in addressing imbalanced data within fault diagnosis.
The robust feature extraction capability of deep learning (DL) has rendered it a popular choice for fault diagnosis in various industries. However, the accuracy of fault diagnosis models is often compromised due to an imbalance between normal and faulty data during the production process. To address this issue, this study proposes an unbalanced bearing fault sample diagnostic method based on a cycle generative adversarial network, using vibration fault diagnosis of rotating machinery rolling bearings as an illustrative example for validation purposes. The method utilizes continuous wavelet transform (CWT) to convert the original vibration signal into time-frequency maps, which are then utilized as input for the cycle generative adversarial network. To address issues related to unstable training and timely convergence of the model, spectral normalization and weight decay techniques are introduced, thereby enhancing the performance of the improved cycle generative adversarial network in generating additional fault samples. Finally, the Swin Transformer model is employed for fault diagnosis and compared with other methods such as Random Forest (RF), Sparse Autoencoder (SAE), Support Vector Machine (SVM), and Convolutional Neural Network (CNN). The diagnostic experiment encompasses sample expansion, imbalanced proportion between normal and faulty samples, as well as scenarios involving limited sample sizes. Experimental results demonstrate that the proposed method can generate higher quality synthetic samples when training data is scarce; moreover, utilizing the Swin Transformer for diagnosis yields superior accuracy compared to alternative approaches such as RF, SAE, SVM, or CNN models – thus showcasing significant potential in addressing imbalanced data within fault diagnosis.
Preparation and photocatalytic performance of GQD doped alkali-exfoliated g-C3N4 composite materials
Abstract:
Graphitic carbon nitride (g-C3N4), as a metal-free two-dimensional nanomaterial, has a broad prospect for photocatalysis. In this paper, alkali-exfoliated g-C3N4 has been modified by sodium tert-butoxide from bulk g-C3N4, where a family of composite materials has been prepared by tuning the dosage of graphene quantum dot (GQD). Simultaneously, the morphology, structure, and composition of these samples have been fully confirmed by XRD, FT-IR, SEM, UV-Vis, PL, and other characterization methods, as well as the photodegradation of methylene blue. The results indicate that compared with bulk g-C3N4, alkali-exfoliated g-C3N4 can shorten the carrier transport distance, and the introduction of GQD can effectively improve the recombination of electron-hole pairs. When the doping amount of GQD is 5.0%, the photocatalytic degradation efficiency of methylene blue reaches 90.9%. In addition, the active groups of photodegradation have been identified as?O2- and h+ by free radical trapping experiments, and a possible methylene blue light catalytic mechanism is proposed.
Graphitic carbon nitride (g-C3N4), as a metal-free two-dimensional nanomaterial, has a broad prospect for photocatalysis. In this paper, alkali-exfoliated g-C3N4 has been modified by sodium tert-butoxide from bulk g-C3N4, where a family of composite materials has been prepared by tuning the dosage of graphene quantum dot (GQD). Simultaneously, the morphology, structure, and composition of these samples have been fully confirmed by XRD, FT-IR, SEM, UV-Vis, PL, and other characterization methods, as well as the photodegradation of methylene blue. The results indicate that compared with bulk g-C3N4, alkali-exfoliated g-C3N4 can shorten the carrier transport distance, and the introduction of GQD can effectively improve the recombination of electron-hole pairs. When the doping amount of GQD is 5.0%, the photocatalytic degradation efficiency of methylene blue reaches 90.9%. In addition, the active groups of photodegradation have been identified as?O2- and h+ by free radical trapping experiments, and a possible methylene blue light catalytic mechanism is proposed.
Abstract:
Autonomous driving technology can improve travel safety and shorten travel time. Reasonable route planning directly affects the safety and comfort of autonomous vehicles. To address the issues of low search efficiency and non-compliance with vehicle kinematic constraints in traditional route planning algorithms for autonomous vehicles, improvements have been made. By replacing the local target points of the dynamic window method with key nodes of the improved ant colony algorithm for route planning, and adding a target distance evaluation sub-function to the evaluation function of the dynamic window method, the efficiency and smoothness of route planning are improved. Additionally, a path replanning method is used to address global path failure, allowing vehicles to escape obstacles and meet safety requirements for route planning. The improved ant colony algorithm uses the location information of start and end points to distribute initial pheromones unevenly, reducing time consumption in the initial search phase. By maintaining the global optimal path and increasing the concentration of pheromones on excellent local paths, the information update mechanism is optimized to accelerate path exploration efficiency. Secondary optimization of planned paths reduces redundancy in nodes and turning points, thus reducing path length. Simulation results show that the integrated algorithm proposed in this paper provides better performance in terms of distance, smoothness, and convergence compared to traditional route planning algorithms, and meets the requirements for safe driving of autonomous vehicles.
Autonomous driving technology can improve travel safety and shorten travel time. Reasonable route planning directly affects the safety and comfort of autonomous vehicles. To address the issues of low search efficiency and non-compliance with vehicle kinematic constraints in traditional route planning algorithms for autonomous vehicles, improvements have been made. By replacing the local target points of the dynamic window method with key nodes of the improved ant colony algorithm for route planning, and adding a target distance evaluation sub-function to the evaluation function of the dynamic window method, the efficiency and smoothness of route planning are improved. Additionally, a path replanning method is used to address global path failure, allowing vehicles to escape obstacles and meet safety requirements for route planning. The improved ant colony algorithm uses the location information of start and end points to distribute initial pheromones unevenly, reducing time consumption in the initial search phase. By maintaining the global optimal path and increasing the concentration of pheromones on excellent local paths, the information update mechanism is optimized to accelerate path exploration efficiency. Secondary optimization of planned paths reduces redundancy in nodes and turning points, thus reducing path length. Simulation results show that the integrated algorithm proposed in this paper provides better performance in terms of distance, smoothness, and convergence compared to traditional route planning algorithms, and meets the requirements for safe driving of autonomous vehicles.
Abstract:
The day-ahead economic dispatch model based on the standard AC power flow equation is beneficial to obtaining accurate solutions. However, it is a nonlinear programming model and has deficiencies in model convergence and computational efficiency, which limits its application in actual power systems. In this context, an economic dispatch model based on improved convex hull theory is proposed to achieve the efficient and accurate solution. Firstly, the day-ahead economic dispatch model based on the standard AC power flow equation is introduced. Secondly, the convex hull theory and the convex hull equation of nonlinear functions are introduced, and then a day-ahead economic dispatch model based on the improved convex hull theory is proposed. Finally, the IEEE 14-bus system, the 118-bus system and a provincial actual power system in China are employed for simulation analysis. The simulation results illustrate that the proposed strategy in this paper can significantly improve the computational efficiency of the model and reduce the relaxation error, which has the higher practical application value.
The day-ahead economic dispatch model based on the standard AC power flow equation is beneficial to obtaining accurate solutions. However, it is a nonlinear programming model and has deficiencies in model convergence and computational efficiency, which limits its application in actual power systems. In this context, an economic dispatch model based on improved convex hull theory is proposed to achieve the efficient and accurate solution. Firstly, the day-ahead economic dispatch model based on the standard AC power flow equation is introduced. Secondly, the convex hull theory and the convex hull equation of nonlinear functions are introduced, and then a day-ahead economic dispatch model based on the improved convex hull theory is proposed. Finally, the IEEE 14-bus system, the 118-bus system and a provincial actual power system in China are employed for simulation analysis. The simulation results illustrate that the proposed strategy in this paper can significantly improve the computational efficiency of the model and reduce the relaxation error, which has the higher practical application value.
Abstract:
To improve the accuracy of State of Health (SOH) monitoring for lithium-ion batteries and prevent potential failures, this paper proposes a joint model based on Convolutional Neural Network (CNN) with enhanced Bidirectional Long Short-Term Memory (BiLSTM) for predicting battery SOH. Firstly, five highly correlated health factors related to battery aging are extracted as SOH features from the publicly available NASA battery dataset. Secondly, a joint network model is constructed where the CNN layer extracts spatial features from the battery charge/discharge characteristic data, the BiLSTM layer captures correlations among different features in the data, and an attention mechanism allocates weights to input variables. Finally, training and prediction are performed on both individual battery datasets and datasets of similar batteries. Simulation results demonstrate that the joint model estimates lithium battery SOH with an average absolute error distribution within 0.015, maximum absolute error and RMAE within 0.01, indicating high accuracy and good generality.
To improve the accuracy of State of Health (SOH) monitoring for lithium-ion batteries and prevent potential failures, this paper proposes a joint model based on Convolutional Neural Network (CNN) with enhanced Bidirectional Long Short-Term Memory (BiLSTM) for predicting battery SOH. Firstly, five highly correlated health factors related to battery aging are extracted as SOH features from the publicly available NASA battery dataset. Secondly, a joint network model is constructed where the CNN layer extracts spatial features from the battery charge/discharge characteristic data, the BiLSTM layer captures correlations among different features in the data, and an attention mechanism allocates weights to input variables. Finally, training and prediction are performed on both individual battery datasets and datasets of similar batteries. Simulation results demonstrate that the joint model estimates lithium battery SOH with an average absolute error distribution within 0.015, maximum absolute error and RMAE within 0.01, indicating high accuracy and good generality.
Abstract:
Distributed computing technology for real-time data in power grid regulation is considered to be one of the important directions in the recent development of the power industry. It involves the distribution of computing tasks to multiple computing nodes for parallel execution, aiming to handle large-scale real-time data in power grid regulation. The current application status of distributed computing technology in real-time data for power grid regulation was comprehensively analyzed in this article, and its future development trends are explored. Firstly, the characteristics and processing methods of real-time data in power grids are discussed ; then, the current state of research on distributed computing technology is examined, along with its application in the context of real-time data in power grids, and an exploration of the development of emerging distributed computing technologies and their integration into power grid applications is presented; finally, an open discussion and future outlook for the distributed computing technology of real-time data in power grid regulation are provided, based on the current research status and trends.
Distributed computing technology for real-time data in power grid regulation is considered to be one of the important directions in the recent development of the power industry. It involves the distribution of computing tasks to multiple computing nodes for parallel execution, aiming to handle large-scale real-time data in power grid regulation. The current application status of distributed computing technology in real-time data for power grid regulation was comprehensively analyzed in this article, and its future development trends are explored. Firstly, the characteristics and processing methods of real-time data in power grids are discussed ; then, the current state of research on distributed computing technology is examined, along with its application in the context of real-time data in power grids, and an exploration of the development of emerging distributed computing technologies and their integration into power grid applications is presented; finally, an open discussion and future outlook for the distributed computing technology of real-time data in power grid regulation are provided, based on the current research status and trends.
Abstract:
A bearing fault diagnosis method based on recurrence analysis and Stacking ensemble learning was proposed to effectively extract nonlinear information from rolling bearing signals and improve diagnostic accuracy. Firstly, the nonlinear information in bearing signals was mapped to a two-dimensional recurrence plot using recurrence analysis theory. Convolutional neural network and support vector machine models were established from the perspectives of image recognition and recurrence quantification analysis, respectively. Finally, the Stacking method was employed to integrate these two models, leveraging their respective strengths. Experimental results demonstrated that the proposed method significantly improved the classification accuracy of bearing vibration signals and exhibited excellent stability under varying load conditions, providing a reliable solution for bearing fault diagnosis.
A bearing fault diagnosis method based on recurrence analysis and Stacking ensemble learning was proposed to effectively extract nonlinear information from rolling bearing signals and improve diagnostic accuracy. Firstly, the nonlinear information in bearing signals was mapped to a two-dimensional recurrence plot using recurrence analysis theory. Convolutional neural network and support vector machine models were established from the perspectives of image recognition and recurrence quantification analysis, respectively. Finally, the Stacking method was employed to integrate these two models, leveraging their respective strengths. Experimental results demonstrated that the proposed method significantly improved the classification accuracy of bearing vibration signals and exhibited excellent stability under varying load conditions, providing a reliable solution for bearing fault diagnosis.
Abstract:
Artificial Intelligence Generated Content (AIGC) technology can provide various types of information generation services for humans. The current urgent issue is how to accurately evaluate the quality of AIGC. This study conducts an in-depth research on the quality of images generated by large models and their evaluation metrics. Firstly, it summarizes the common methods for evaluating AIGC from a technical perspective, such as deep learning methods and computer vision methods. It focuses on introducing the metrics used in the evaluation methods, including accuracy, relevance, consistency, and interpretability, and analyzes their performance in evaluating different types of generated content. Then, to demonstrate the practical application of these evaluation metrics, this study conducts an evaluation experiment on the images generated by ERNIE Bot as an example. Objective evaluation of the generated images is carried out using quantitative metrics such as histograms and noise counts. Subjective evaluation is performed by assessing the overall coordination and aesthetic appeal of the images. Finally, by comparing the results of objective and subjective evaluations, this study screens out highly reliable metrics for evaluating the quality of AIGC images, such as color bias, noise count, and psychological expectations. This research not only provides a theoretical reference for evaluating the quality of AIGC, but also proves the effectiveness and reliability of using both objective and subjective evaluation metrics for AIGC product evaluation through experimental results. It has certain reference significance for obtaining feedback on model performance and optimizing performance.
Artificial Intelligence Generated Content (AIGC) technology can provide various types of information generation services for humans. The current urgent issue is how to accurately evaluate the quality of AIGC. This study conducts an in-depth research on the quality of images generated by large models and their evaluation metrics. Firstly, it summarizes the common methods for evaluating AIGC from a technical perspective, such as deep learning methods and computer vision methods. It focuses on introducing the metrics used in the evaluation methods, including accuracy, relevance, consistency, and interpretability, and analyzes their performance in evaluating different types of generated content. Then, to demonstrate the practical application of these evaluation metrics, this study conducts an evaluation experiment on the images generated by ERNIE Bot as an example. Objective evaluation of the generated images is carried out using quantitative metrics such as histograms and noise counts. Subjective evaluation is performed by assessing the overall coordination and aesthetic appeal of the images. Finally, by comparing the results of objective and subjective evaluations, this study screens out highly reliable metrics for evaluating the quality of AIGC images, such as color bias, noise count, and psychological expectations. This research not only provides a theoretical reference for evaluating the quality of AIGC, but also proves the effectiveness and reliability of using both objective and subjective evaluation metrics for AIGC product evaluation through experimental results. It has certain reference significance for obtaining feedback on model performance and optimizing performance.
Abstract:
The classification and accurate recognition of transformer operating states are very important for the stable operation of transformers and the safe power supply of power systems. At present, there are still some problems in this kind of research, such as less attention to the relevant data of transformer loads, high complexity of mechanism models and unclear correlation between oil temperature data and overload state. Therefore, this paper proposes an improved hybrid model that combines the hunter-prey optimization (HPO) algorithm with the efficient channel attention (ECA) module, applied to convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) neural networks. It is used to classify transformer operating state and identify overload fault. The data of a main transformer containing nine characteristics related to transformer load are selected as samples, and the load state category is selected through K-Means ++ clustering and transformer normal periodic load analysis. The parameters of the hybrid model were optimized by HPO to improve the performance and generalization ability. Through pre-processing and feature extraction of transformer load data, the improved model is used to recognize the load state accurately. The experimental results show that the recognition accuracy of this method can reach 99.24%, and good results are obtained in the task of transformer operating state classification.
The classification and accurate recognition of transformer operating states are very important for the stable operation of transformers and the safe power supply of power systems. At present, there are still some problems in this kind of research, such as less attention to the relevant data of transformer loads, high complexity of mechanism models and unclear correlation between oil temperature data and overload state. Therefore, this paper proposes an improved hybrid model that combines the hunter-prey optimization (HPO) algorithm with the efficient channel attention (ECA) module, applied to convolutional neural networks (CNN) and bidirectional long short-term memory (BiLSTM) neural networks. It is used to classify transformer operating state and identify overload fault. The data of a main transformer containing nine characteristics related to transformer load are selected as samples, and the load state category is selected through K-Means ++ clustering and transformer normal periodic load analysis. The parameters of the hybrid model were optimized by HPO to improve the performance and generalization ability. Through pre-processing and feature extraction of transformer load data, the improved model is used to recognize the load state accurately. The experimental results show that the recognition accuracy of this method can reach 99.24%, and good results are obtained in the task of transformer operating state classification.
Abstract:
In order to enable picking robots to quickly and accurately detect apple fruits with varying levels of maturity in complex orchard environments (such as different lighting conditions, leaf occlusions, dense apple clusters, and ultra-long viewing distances), this study proposes an apple fruit detection model based on an improved YOLOv8. First, the EMA attention mechanism module is integrated into the YOLOv8 model, making the model more focused on the region of interest for fruit detection, suppressing general feature information such as background and branch leaf occlusion, and improving the detection accuracy of occluded fruits. Second, the original C2f module is reconstructed using a more efficient three-branch DWR module for feature extraction, which enhances the small object detection capability through multi-scale feature fusion methods. Simultaneously, based on the DAMO-YOLO idea, the original YOLOv8 neck is reconstructed to achieve efficient fusion of high-level semantics and low-level spatial features. Finally, the model is optimized using the Inner-SIoU loss function to improve the recognition accuracy. In the complex orchard environment, using apples as the detection target, the experimental results show that the proposed algorithm achieves P, R, mAP50, mAP50-95, and F1 of 86.1%, 89.2%, 94.0%, 64.4%, and 87.6% on the test set, respectively. The improved algorithm outperforms the original model in most indicators, and demonstrates excellent robustness through comparative experiments with different numbers of fruits. This provides practical application value for addressing the precise identification problem of fruit picking robots in complex environments.
In order to enable picking robots to quickly and accurately detect apple fruits with varying levels of maturity in complex orchard environments (such as different lighting conditions, leaf occlusions, dense apple clusters, and ultra-long viewing distances), this study proposes an apple fruit detection model based on an improved YOLOv8. First, the EMA attention mechanism module is integrated into the YOLOv8 model, making the model more focused on the region of interest for fruit detection, suppressing general feature information such as background and branch leaf occlusion, and improving the detection accuracy of occluded fruits. Second, the original C2f module is reconstructed using a more efficient three-branch DWR module for feature extraction, which enhances the small object detection capability through multi-scale feature fusion methods. Simultaneously, based on the DAMO-YOLO idea, the original YOLOv8 neck is reconstructed to achieve efficient fusion of high-level semantics and low-level spatial features. Finally, the model is optimized using the Inner-SIoU loss function to improve the recognition accuracy. In the complex orchard environment, using apples as the detection target, the experimental results show that the proposed algorithm achieves P, R, mAP50, mAP50-95, and F1 of 86.1%, 89.2%, 94.0%, 64.4%, and 87.6% on the test set, respectively. The improved algorithm outperforms the original model in most indicators, and demonstrates excellent robustness through comparative experiments with different numbers of fruits. This provides practical application value for addressing the precise identification problem of fruit picking robots in complex environments.
Abstract:
Aquatic vegetation is an important indicator of river health, and it is important to carry out dynamic monitoring of river aquatic vegetation. However, both traditional field surveys and satellite remote sensing have obvious limitations in monitoring river aquatic vegetation. In this study, a UAV (Unmanned Aerial Vehicle)-borne high-pixel Aerospace Shuwei KP-6 multispectral camera was used to acquire multispectral images of the Baigou River in Gu'an County, Hebei Province. Aquatic vegetation species such as reed, Nymphoides Peltata, and Ceratophyllum demersum L. were classified in the Baigou River based on three kinds of supervised classifications, and the maximum-likelihood method was examined to have the highest precision, with an overall precision of 92.8% and the Kappa coefficient of 0.91. In addition, the high-pixel Aerospace Shuwei KP-6 camera was compared with a variety of high-resolution satellites and a number of UAV-borne multispectral cameras, and was found to have the advantages of high spatial resolution, wide coverage and flexible acquisition in river aquatic vegetation monitoring. The present study can provide a useful reference for the application of UAV-borne multispectral remote sensing in riverine aquatic vegetation monitoring.
Aquatic vegetation is an important indicator of river health, and it is important to carry out dynamic monitoring of river aquatic vegetation. However, both traditional field surveys and satellite remote sensing have obvious limitations in monitoring river aquatic vegetation. In this study, a UAV (Unmanned Aerial Vehicle)-borne high-pixel Aerospace Shuwei KP-6 multispectral camera was used to acquire multispectral images of the Baigou River in Gu'an County, Hebei Province. Aquatic vegetation species such as reed, Nymphoides Peltata, and Ceratophyllum demersum L. were classified in the Baigou River based on three kinds of supervised classifications, and the maximum-likelihood method was examined to have the highest precision, with an overall precision of 92.8% and the Kappa coefficient of 0.91. In addition, the high-pixel Aerospace Shuwei KP-6 camera was compared with a variety of high-resolution satellites and a number of UAV-borne multispectral cameras, and was found to have the advantages of high spatial resolution, wide coverage and flexible acquisition in river aquatic vegetation monitoring. The present study can provide a useful reference for the application of UAV-borne multispectral remote sensing in riverine aquatic vegetation monitoring.
Abstract:
To fully extract the frequency information and spatial topology information of multi-channel EEG signals, this paper introduces an EEG emotion recognition model——CBAM-CapsNet, which utilizes a capsule network with a hybrid attention module. Firstly, EEG signals from different frequency bands are acquired, and their differential entropy features are extracted. Secondly, these features are mapped into a three-dimensional compact feature matrix based on spatial lead distribution. Finally, the three-dimensional feature matrix is processed through a capsule network with a hybrid module (CBAM-CapsNet) for training and prediction. Experimental results indicate that higher frequency bands have a more significant impact on emotion recognition, and the use of four-frequency band three-dimensional matrix can significantly enhance the accuracy of emotion recognition. CBAM-CapsNet achieved a binary classification accuracy of 95.42% and 95.52% for Arousal and Valence, respectively, on the DEAP dataset, a four-class accuracy of 95.00% for Arousal_Valence combined on the DEAP dataset, and a three-class accuracy of 93.19% on the SEED dataset. Compared to existing mainstream brain emotion recognition models based on deep learning, the accuracy is significantly improved.
To fully extract the frequency information and spatial topology information of multi-channel EEG signals, this paper introduces an EEG emotion recognition model——CBAM-CapsNet, which utilizes a capsule network with a hybrid attention module. Firstly, EEG signals from different frequency bands are acquired, and their differential entropy features are extracted. Secondly, these features are mapped into a three-dimensional compact feature matrix based on spatial lead distribution. Finally, the three-dimensional feature matrix is processed through a capsule network with a hybrid module (CBAM-CapsNet) for training and prediction. Experimental results indicate that higher frequency bands have a more significant impact on emotion recognition, and the use of four-frequency band three-dimensional matrix can significantly enhance the accuracy of emotion recognition. CBAM-CapsNet achieved a binary classification accuracy of 95.42% and 95.52% for Arousal and Valence, respectively, on the DEAP dataset, a four-class accuracy of 95.00% for Arousal_Valence combined on the DEAP dataset, and a three-class accuracy of 93.19% on the SEED dataset. Compared to existing mainstream brain emotion recognition models based on deep learning, the accuracy is significantly improved.
Abstract:
With the deep integration of artificial intelligence technology and wireless communication, semantic communication technology emerges as a vital mode focusing on semantic-level information transmission and interaction, thereby significantly enhancing communication accuracy and reliability. In the scenarios of low latency and high traffic density communication applications, semantic communication technology surpasses traditional syntactic-level communication grounded in classical information theory, presenting a new paradigm in wireless communication and expanding the application scope of modern communication technology. Nonetheless, the development of semantic communication technology is still in its infancy, and the security issues it faces in the application process have not been thoroughly researched and comprehensively analyzed. To advance the development and implementation of semantic communication technology, the paper first provides an overview of various security threats in semantic communication systems; then, it details the research status of model security and data security in semantic communication systems; finally, it summarizes the challenges and future trends in the security research of semantic communication.
With the deep integration of artificial intelligence technology and wireless communication, semantic communication technology emerges as a vital mode focusing on semantic-level information transmission and interaction, thereby significantly enhancing communication accuracy and reliability. In the scenarios of low latency and high traffic density communication applications, semantic communication technology surpasses traditional syntactic-level communication grounded in classical information theory, presenting a new paradigm in wireless communication and expanding the application scope of modern communication technology. Nonetheless, the development of semantic communication technology is still in its infancy, and the security issues it faces in the application process have not been thoroughly researched and comprehensively analyzed. To advance the development and implementation of semantic communication technology, the paper first provides an overview of various security threats in semantic communication systems; then, it details the research status of model security and data security in semantic communication systems; finally, it summarizes the challenges and future trends in the security research of semantic communication.
Abstract:
Aiming at the problem of extracting cavern surface deformation by three-dimensional laser scanning dense point clouds, a method of cavern surface deformation monitoring based on Multiscale Model-to-Model Cloud Comparison(M3C2) and improved Alpha Shapes algorithm is proposed. Firstly, the two phase surface point cloud data are registered, and the improved Alpha Shapes algorithm is used to identify the outer contour point cloud. After the fine registration of the two phase outer contour point clouds, the M3C2 algorithm is used to calculate the deformation value of each point, and finally the continuous deformation region is extracted by distance clustering. The experimental results show that the proposed method can effectively eliminate the points at the small furrows and the points affected by the mixed pixels. The removal rates of the point cloud in the two phases are 14.17% and 13.52% within 10m of the cavern section to the scanner, respectively, 6.25% and 6.42% within 70m. This method can accurately and efficiently extract the deformation region of the cavern surface with more than 2 times the registration error.
Aiming at the problem of extracting cavern surface deformation by three-dimensional laser scanning dense point clouds, a method of cavern surface deformation monitoring based on Multiscale Model-to-Model Cloud Comparison(M3C2) and improved Alpha Shapes algorithm is proposed. Firstly, the two phase surface point cloud data are registered, and the improved Alpha Shapes algorithm is used to identify the outer contour point cloud. After the fine registration of the two phase outer contour point clouds, the M3C2 algorithm is used to calculate the deformation value of each point, and finally the continuous deformation region is extracted by distance clustering. The experimental results show that the proposed method can effectively eliminate the points at the small furrows and the points affected by the mixed pixels. The removal rates of the point cloud in the two phases are 14.17% and 13.52% within 10m of the cavern section to the scanner, respectively, 6.25% and 6.42% within 70m. This method can accurately and efficiently extract the deformation region of the cavern surface with more than 2 times the registration error.
Abstract:
Aiming at the problem of frequent failures for the wind turbines working in harsh environment for a long time, an anomaly warning method with convolutional neural network (CNN) and gate recurrent unit (GRU) with attention mechanism is proposed. Firstly, the clustering fast search and find of density peaks (CFSFDP) algorithm and the local outlier factor (LOF) algorithm are jointly employed to preprocess the abnormal data in the wind turbine condition supervisory control and data acquisition (SCADA) system. Then the input and output parameters of the model are selected by evaluating the feature importances with the extreme gradient boosting (XGBoost) algorithm. The selected parameters are then input into the CNN-GRU-Attention model for training and prediction to establish the prediction model of the wind turbine under normal operation conditions. Based on this model, the state evaluation index of wind turbine is established by using the time-lapse sliding window, and the upper warning threshold is determined by the interval estimation method in statistics to realize the abnormal condition early warning. Finally, fault warning experiments are carried out with the real historical fault data of a wind power unit, which show that the CNN-GRU-Attention model can effectively identify and warn the abnormal working conditions in advance to ensure the stable operation of the wind turbine.
Aiming at the problem of frequent failures for the wind turbines working in harsh environment for a long time, an anomaly warning method with convolutional neural network (CNN) and gate recurrent unit (GRU) with attention mechanism is proposed. Firstly, the clustering fast search and find of density peaks (CFSFDP) algorithm and the local outlier factor (LOF) algorithm are jointly employed to preprocess the abnormal data in the wind turbine condition supervisory control and data acquisition (SCADA) system. Then the input and output parameters of the model are selected by evaluating the feature importances with the extreme gradient boosting (XGBoost) algorithm. The selected parameters are then input into the CNN-GRU-Attention model for training and prediction to establish the prediction model of the wind turbine under normal operation conditions. Based on this model, the state evaluation index of wind turbine is established by using the time-lapse sliding window, and the upper warning threshold is determined by the interval estimation method in statistics to realize the abnormal condition early warning. Finally, fault warning experiments are carried out with the real historical fault data of a wind power unit, which show that the CNN-GRU-Attention model can effectively identify and warn the abnormal working conditions in advance to ensure the stable operation of the wind turbine.
Abstract:
The coastal zone serves as a critical interface for the fixation and storage of organic carbon at the land-sea boundary, exhibiting significant spatiotemporal characteristics and patterns in organic carbon formation over millennia of cultivated soil development. This study aimed to analyze the vertical distribution characteristics and temporal evolution patterns of soil organic carbon (SOC) of sandy-silty coastal zones of the South Yellow Sea over millennia. Twenty one-meter-deep soil profiles were collected based on the position of the coastlines from Sheyang in the north and Dongtai in the middle of the study area. The spatiotemporal distribution characteristics of SOC mass fraction (SOCC), density (SOCD), and stock (SOCS) were analyzed, along with their correlation with soil physicochemical properties. Results indicated: 1) Over millennia, SOCC in coastal cultivated soil profiles ranged from 1.00 to 24.44 g·kg-1, while SOCD varied within 0.13-2.78 g·m-2, with both SOCC and SOCD decreasing with depth in the two regions. 2) The temporal accumulation functions of SOCS differed between the two regions, with SOCS in Sheyang showing a linear relationship with soil age, while in Dongtai, SOCS exhibited a logarithmic function, primarily due to differences in soil texture. 3) Significant correlations were found between coastal cultivated SOC and soil physicochemical properties, with Sheyang showing higher correlations between SOC and bulk density, total salinity, whereas Dongtai exhibited higher correlations with soil texture. This study provides important insights into the dynamic characteristics and long-term storage mechanisms of coastal cultivated soil carbon pools, offering valuable references for understanding these systems.
The coastal zone serves as a critical interface for the fixation and storage of organic carbon at the land-sea boundary, exhibiting significant spatiotemporal characteristics and patterns in organic carbon formation over millennia of cultivated soil development. This study aimed to analyze the vertical distribution characteristics and temporal evolution patterns of soil organic carbon (SOC) of sandy-silty coastal zones of the South Yellow Sea over millennia. Twenty one-meter-deep soil profiles were collected based on the position of the coastlines from Sheyang in the north and Dongtai in the middle of the study area. The spatiotemporal distribution characteristics of SOC mass fraction (SOCC), density (SOCD), and stock (SOCS) were analyzed, along with their correlation with soil physicochemical properties. Results indicated: 1) Over millennia, SOCC in coastal cultivated soil profiles ranged from 1.00 to 24.44 g·kg-1, while SOCD varied within 0.13-2.78 g·m-2, with both SOCC and SOCD decreasing with depth in the two regions. 2) The temporal accumulation functions of SOCS differed between the two regions, with SOCS in Sheyang showing a linear relationship with soil age, while in Dongtai, SOCS exhibited a logarithmic function, primarily due to differences in soil texture. 3) Significant correlations were found between coastal cultivated SOC and soil physicochemical properties, with Sheyang showing higher correlations between SOC and bulk density, total salinity, whereas Dongtai exhibited higher correlations with soil texture. This study provides important insights into the dynamic characteristics and long-term storage mechanisms of coastal cultivated soil carbon pools, offering valuable references for understanding these systems.
Abstract:
The sampling-loop abnormality of smart substations has the characteristics of concealment, instantaneousness, and instability, which is difficult to be found. In this regard, a smart substations sampling-loop abnormality warning scheme based on homology recording data is proposed. First, whether the sampling loop is abnormal is judged by reasonably setting the early warning thresholds and waveform-analysis criteria. Second, based on the information of the equipment operation and maintenance manual related to the sampling loop, the knowledge map of the operation and maintenance of the sampling loop is constructed to realize the abnormal decision-making. Finally, a case study is carried out by taking the actual defect of a substation as an example. The results show that the scheme is effective in finding the abnormality of the sampling loops, and the constructed knowledge graph can be used as a reference for O&M personnel to dispose faults, which greatly alleviates the manual pressure.
The sampling-loop abnormality of smart substations has the characteristics of concealment, instantaneousness, and instability, which is difficult to be found. In this regard, a smart substations sampling-loop abnormality warning scheme based on homology recording data is proposed. First, whether the sampling loop is abnormal is judged by reasonably setting the early warning thresholds and waveform-analysis criteria. Second, based on the information of the equipment operation and maintenance manual related to the sampling loop, the knowledge map of the operation and maintenance of the sampling loop is constructed to realize the abnormal decision-making. Finally, a case study is carried out by taking the actual defect of a substation as an example. The results show that the scheme is effective in finding the abnormality of the sampling loops, and the constructed knowledge graph can be used as a reference for O&M personnel to dispose faults, which greatly alleviates the manual pressure.
Abstract:
Aiming at the safety and comfort problems of intelligent vehicles in highway lane-changing scenarios, a trajectory planning method based on safety potential field and polynomial lane-changing model is proposed. Firstly, the motion of the vehicle is decoupled into horizontal and vertical dimensions in the Frenet coordinate system, and the horizontal d-t trajectory cluster and the longitudinal s-t trajectory cluster of the vehicle are generated by the fifth-order polynomial and the fourth-order polynomial. Secondly, in order to improve the efficiency of the algorithm, a trajectory evaluation index including acceleration, acceleration change rate and curvature is designed according to the vehicle dynamics characteristics, and the candidate trajectory is obtained after the initial screening of the trajectory cluster. Finally, based on the safety potential field theory and the concept of minimum driving safety distance, a trajectory evaluation function including safety, comfort and efficiency is established to select the optimal trajectory for the candidate trajectory and complete the simulation verification. The algorithm is simulated and verified by building a high-speed two-lane curve model and designing different lane changing scenarios of uniform traffic flow and variable-speed traffic flow. The results show that in the process of lane changing, the collision risk value between the self-vehicle and each obstacle vehicle is always less than the critical value of collision risk, which ensures the safety of lane changing. Under different driving conditions, the acceleration, acceleration change rate and trajectory curvature of the self-driving car are all less than the threshold, indicating that the lane change trajectory planning algorithm can ensure the comfort and smoothness of the lane change trajectory of the self-driving car in a variety of obstacle traffic flows.
Aiming at the safety and comfort problems of intelligent vehicles in highway lane-changing scenarios, a trajectory planning method based on safety potential field and polynomial lane-changing model is proposed. Firstly, the motion of the vehicle is decoupled into horizontal and vertical dimensions in the Frenet coordinate system, and the horizontal d-t trajectory cluster and the longitudinal s-t trajectory cluster of the vehicle are generated by the fifth-order polynomial and the fourth-order polynomial. Secondly, in order to improve the efficiency of the algorithm, a trajectory evaluation index including acceleration, acceleration change rate and curvature is designed according to the vehicle dynamics characteristics, and the candidate trajectory is obtained after the initial screening of the trajectory cluster. Finally, based on the safety potential field theory and the concept of minimum driving safety distance, a trajectory evaluation function including safety, comfort and efficiency is established to select the optimal trajectory for the candidate trajectory and complete the simulation verification. The algorithm is simulated and verified by building a high-speed two-lane curve model and designing different lane changing scenarios of uniform traffic flow and variable-speed traffic flow. The results show that in the process of lane changing, the collision risk value between the self-vehicle and each obstacle vehicle is always less than the critical value of collision risk, which ensures the safety of lane changing. Under different driving conditions, the acceleration, acceleration change rate and trajectory curvature of the self-driving car are all less than the threshold, indicating that the lane change trajectory planning algorithm can ensure the comfort and smoothness of the lane change trajectory of the self-driving car in a variety of obstacle traffic flows.
Abstract:
With the rapid development of the power industry, the monitoring of the state of power generation units has garnered widespread attention. In order to effectively detect whether wind turbine blades are iced, this paper proposes a wind turbine blade icing detection method based on SCADA (Supervisory Control and Data Acquisition) image data enhancement. The method initially reconstructs features based on the mechanism of icing, considering both data relevance and feature importance for feature selection. Subsequently, SCADA data is transformed into a two-dimensional image format to meet the input requirements of two-dimensional neural network models. On this basis, an optimized CycleGAN algorithm is used to generate more adaptable image data, aiming to address the issue of data category imbalance and significantly enhance the generalization ability of the model. This study selected WT15 data as the training set and WT21 data as the test set. The experimental results demonstrate that the feature selection method proposed in this paper leads to an increase of 4.79 percentage points in model accuracy and 2.64 percentage points in F1 score, compared to feature selection using the XGBoost model. In comparison to the original CycleGAN model, the improved model achieved an increase in accuracy of 6.78 percentage points and an improvement in the F1 score of 6.91 percentage points. These findings indicate that the method proposed in this paper offers a significant advantage in improving the model's accuracy and generalisation ability.
With the rapid development of the power industry, the monitoring of the state of power generation units has garnered widespread attention. In order to effectively detect whether wind turbine blades are iced, this paper proposes a wind turbine blade icing detection method based on SCADA (Supervisory Control and Data Acquisition) image data enhancement. The method initially reconstructs features based on the mechanism of icing, considering both data relevance and feature importance for feature selection. Subsequently, SCADA data is transformed into a two-dimensional image format to meet the input requirements of two-dimensional neural network models. On this basis, an optimized CycleGAN algorithm is used to generate more adaptable image data, aiming to address the issue of data category imbalance and significantly enhance the generalization ability of the model. This study selected WT15 data as the training set and WT21 data as the test set. The experimental results demonstrate that the feature selection method proposed in this paper leads to an increase of 4.79 percentage points in model accuracy and 2.64 percentage points in F1 score, compared to feature selection using the XGBoost model. In comparison to the original CycleGAN model, the improved model achieved an increase in accuracy of 6.78 percentage points and an improvement in the F1 score of 6.91 percentage points. These findings indicate that the method proposed in this paper offers a significant advantage in improving the model's accuracy and generalisation ability.
Abstract:
RGB-Infrared person re-identification is a challenging task aimed at matching person images between visible and infrared modalities, playing a crucial role in criminal investigation and intelligent surveillance. To address the issue of weak feature extraction capability for fine-grained features in current cross-modal person re-identification tasks, this paper proposes a person re-identification model based on fused attention and feature enhancement. Firstly, automatic data augmentation technique is employed to mitigate perspective and scale differences among different cameras. Secondly, a cross-attention multi-scale Vision Transformer is introduced to generate more discriminative feature representations by processing multi-scale features. Furthermore, channel attention and spatial attention mechanisms are proposed to learn important information for distinguishing features when fusing visible and infrared image features. Finally, a loss function based on adaptive weights is designed, presenting a hard triplet loss to enhance the correlation between each sample and improve the identification capability of visible and infrared images for different persons. Extensive experiments conducted on the SYSU-MM01 and RegDB datasets show that the proposed method achieves mAP scores of 68.05% and 85.19%, respectively, outperforming previous works. Moreover, ablation experiments and comparative analysis validate the superiority and effectiveness of the proposed model.
RGB-Infrared person re-identification is a challenging task aimed at matching person images between visible and infrared modalities, playing a crucial role in criminal investigation and intelligent surveillance. To address the issue of weak feature extraction capability for fine-grained features in current cross-modal person re-identification tasks, this paper proposes a person re-identification model based on fused attention and feature enhancement. Firstly, automatic data augmentation technique is employed to mitigate perspective and scale differences among different cameras. Secondly, a cross-attention multi-scale Vision Transformer is introduced to generate more discriminative feature representations by processing multi-scale features. Furthermore, channel attention and spatial attention mechanisms are proposed to learn important information for distinguishing features when fusing visible and infrared image features. Finally, a loss function based on adaptive weights is designed, presenting a hard triplet loss to enhance the correlation between each sample and improve the identification capability of visible and infrared images for different persons. Extensive experiments conducted on the SYSU-MM01 and RegDB datasets show that the proposed method achieves mAP scores of 68.05% and 85.19%, respectively, outperforming previous works. Moreover, ablation experiments and comparative analysis validate the superiority and effectiveness of the proposed model.
Abstract:
A mathematical model is formulated for the multi-objective software project scheduling problem, aiming to optimize both the project duration and employee satisfaction. The model takes into account practical factors such as the skill level classification of employees and the importance of tasks, and matches important tasks with employees of high skill levels. A hyper-heuristic algorithm based on Q-learning is proposed to solve the model. A global search of the task-employee matrix is performed based on the matrix crossover operator and Jaya operator with random jitter; The local search strategies are designed to reduce the project duration and increase the employee satisfaction by using the problem information; The global search operators, the neighborhood parameter values, and the local search strategies are combined to form eight low-level heuristics; Providing a high-level strategy based on Q-learning that adaptively selects appropriate low-level heuristics for different evolutionary states of the population, based on the historical performance of the low-level heuristics. The experimental results show that the proposed algorithm outperforms the representative algorithms in terms of HVR and IGD on most of the cases.
A mathematical model is formulated for the multi-objective software project scheduling problem, aiming to optimize both the project duration and employee satisfaction. The model takes into account practical factors such as the skill level classification of employees and the importance of tasks, and matches important tasks with employees of high skill levels. A hyper-heuristic algorithm based on Q-learning is proposed to solve the model. A global search of the task-employee matrix is performed based on the matrix crossover operator and Jaya operator with random jitter; The local search strategies are designed to reduce the project duration and increase the employee satisfaction by using the problem information; The global search operators, the neighborhood parameter values, and the local search strategies are combined to form eight low-level heuristics; Providing a high-level strategy based on Q-learning that adaptively selects appropriate low-level heuristics for different evolutionary states of the population, based on the historical performance of the low-level heuristics. The experimental results show that the proposed algorithm outperforms the representative algorithms in terms of HVR and IGD on most of the cases.
Abstract:
To address the issue of inefficient processing of machine vision tasks in industrial environments caused by uneven blurring in images captured in moving scenes, a motion blur image restoration algorithm based on multi-weight adaptive interaction is proposed. Firstly, a multi-strategy feature extraction module is employed to extract shallow and critical texture information from blurred images and smooth noise. Then, a dual-channel adaptive weight extraction module is proposed to capture spatial and pixel weight information from degraded images. Meanwhile, a residual semantic block is constructed to deeply mine the deep semantic information of the image and gradually compensate this information into the network. Finally, a weighted feature fusion module is designed to fuse the multi-spatial weighted features extracted by the network, and multiple loss functions are combined to further improve image quality. The subjective, objective and ablation experimental results of the proposed algorithm in the standard data set show that the SSIM and PSNR indexes in the standard data set reach 0.93 and 31.89, and each module can be well coordinated, which has significant advantages in restoring non-uniform blurred images in moving scenes.
To address the issue of inefficient processing of machine vision tasks in industrial environments caused by uneven blurring in images captured in moving scenes, a motion blur image restoration algorithm based on multi-weight adaptive interaction is proposed. Firstly, a multi-strategy feature extraction module is employed to extract shallow and critical texture information from blurred images and smooth noise. Then, a dual-channel adaptive weight extraction module is proposed to capture spatial and pixel weight information from degraded images. Meanwhile, a residual semantic block is constructed to deeply mine the deep semantic information of the image and gradually compensate this information into the network. Finally, a weighted feature fusion module is designed to fuse the multi-spatial weighted features extracted by the network, and multiple loss functions are combined to further improve image quality. The subjective, objective and ablation experimental results of the proposed algorithm in the standard data set show that the SSIM and PSNR indexes in the standard data set reach 0.93 and 31.89, and each module can be well coordinated, which has significant advantages in restoring non-uniform blurred images in moving scenes.
Abstract:
It is an effective solution to utilize the energy storage unit (ESU) configured inside each wind turbine of the wind farm to absorb the unbalanced power during the DC fault of the wind power integration through a bipolar MMC-HVDC. However, the existing literature only equates the wind farm to a single wind generator (WG) and does not consider the residual capacity difference of each ESU. It is easy to cause the ESU with small residual capacity to be overloaded, while the ESU with large residual capacity still has a large energy storage capacity that is not utilized, resulting in power imbalance during the fault. To address the above problems, this paper proposes a coordinated control strategy for DC fault ride-through based on optimization control of ESUs inside WGs. In this strategy, the variance of state of charge (SOC) is used as an index to quantitatively describe the difference of residual capacity of ESUs, and the maximum decrease rate of SOC variance is taken as the objective function. The residual unbalanced power after the conversion of non-fault pole converter station is optimally allocated to the ESUs within each WG, so as to reduce the difference of residual capacity while ensuring the power balance of the system during the fault period. A model is developed and carried out on the PSCAD/EMTDC simulation platform to compare the proposed power optimization allocation scheme with the traditional average allocation scheme. The results show that the optimal allocation scheme fully developed the power consumption ability of the energy storage system and improves the DC fault ride-through ability of the system.
It is an effective solution to utilize the energy storage unit (ESU) configured inside each wind turbine of the wind farm to absorb the unbalanced power during the DC fault of the wind power integration through a bipolar MMC-HVDC. However, the existing literature only equates the wind farm to a single wind generator (WG) and does not consider the residual capacity difference of each ESU. It is easy to cause the ESU with small residual capacity to be overloaded, while the ESU with large residual capacity still has a large energy storage capacity that is not utilized, resulting in power imbalance during the fault. To address the above problems, this paper proposes a coordinated control strategy for DC fault ride-through based on optimization control of ESUs inside WGs. In this strategy, the variance of state of charge (SOC) is used as an index to quantitatively describe the difference of residual capacity of ESUs, and the maximum decrease rate of SOC variance is taken as the objective function. The residual unbalanced power after the conversion of non-fault pole converter station is optimally allocated to the ESUs within each WG, so as to reduce the difference of residual capacity while ensuring the power balance of the system during the fault period. A model is developed and carried out on the PSCAD/EMTDC simulation platform to compare the proposed power optimization allocation scheme with the traditional average allocation scheme. The results show that the optimal allocation scheme fully developed the power consumption ability of the energy storage system and improves the DC fault ride-through ability of the system.
Abstract:
To explore the response of N2O emissions from greenhouse soil to inoculated plant growth-promoting rhizobacteria (PGPR), four PGPR strains, including Bacillus subtilis subsp. subtilis NRCB002, Stutzerimonas stutzeri NRCB010, Bacillus velezensis NRCB026, and Achromobacter denitrificans YSQ030, were used as test strains. Tomato was used as the test crop, and randomized block design experiments were conducted under greenhouse conditions. The results showed that even under moderately saline stress soil conditions, PGPR inoculation treatment exhibited N2O emission reduction tendency in the first season compared with the CK treatment. In the second season, the treatment with YSQ030 inoculation showed a significant emission reduction effect, reducing soil N2O cumulative emissions by 36%. The four strains of PGPR demonstrated certain promoting effects on the growth and yield of tomato plants. The research results indicated that PGPR played an important role in reducing N2O emissions from greenhouse soil and promoting crop growth.
To explore the response of N2O emissions from greenhouse soil to inoculated plant growth-promoting rhizobacteria (PGPR), four PGPR strains, including Bacillus subtilis subsp. subtilis NRCB002, Stutzerimonas stutzeri NRCB010, Bacillus velezensis NRCB026, and Achromobacter denitrificans YSQ030, were used as test strains. Tomato was used as the test crop, and randomized block design experiments were conducted under greenhouse conditions. The results showed that even under moderately saline stress soil conditions, PGPR inoculation treatment exhibited N2O emission reduction tendency in the first season compared with the CK treatment. In the second season, the treatment with YSQ030 inoculation showed a significant emission reduction effect, reducing soil N2O cumulative emissions by 36%. The four strains of PGPR demonstrated certain promoting effects on the growth and yield of tomato plants. The research results indicated that PGPR played an important role in reducing N2O emissions from greenhouse soil and promoting crop growth.
Abstract:
In the field of Brain-Computer Interfaces (BCI), the recognition of natural hand movements through EEG is crucial for achieving natural and precise human-machine interaction. However, attempts to enhance the generalization ability of models across different subjects using transfer learning are still rare in studies focusing on natural hand movement paradigms. Addressing this issue, this paper conducted EEG experiments with three natural hand movements: grasping, pinching, and twisting. We validated the effectiveness of two transfer learning algorithms, CA-MDM and CA-JDA, on the experimental dataset. The results showed that CA-JDA achieved average accuracies of 60.51%±5.78% and 34.89%±4.42% in binary and four-class tasks, respectively, while CA-MDM performed at 63.88%±4.59% and 35.71%±4.84% in the same tasks, highlighting the advantages of Riemannian space-based classifiers in handling covariance features. This study not only confirms the feasibility of transfer learning in natural hand movement paradigms but also aids in reducing calibration time for BCI systems and implementing natural human-machine interaction strategies.
In the field of Brain-Computer Interfaces (BCI), the recognition of natural hand movements through EEG is crucial for achieving natural and precise human-machine interaction. However, attempts to enhance the generalization ability of models across different subjects using transfer learning are still rare in studies focusing on natural hand movement paradigms. Addressing this issue, this paper conducted EEG experiments with three natural hand movements: grasping, pinching, and twisting. We validated the effectiveness of two transfer learning algorithms, CA-MDM and CA-JDA, on the experimental dataset. The results showed that CA-JDA achieved average accuracies of 60.51%±5.78% and 34.89%±4.42% in binary and four-class tasks, respectively, while CA-MDM performed at 63.88%±4.59% and 35.71%±4.84% in the same tasks, highlighting the advantages of Riemannian space-based classifiers in handling covariance features. This study not only confirms the feasibility of transfer learning in natural hand movement paradigms but also aids in reducing calibration time for BCI systems and implementing natural human-machine interaction strategies.
Abstract:
Extracting valuable landslide geohazard entities from massive texts describing landslide geological hazards is the basis for constructing a landslide geohazard knowledge graph. Based on relevant unstructured text data such as landslide geological hazard exploration report documents, this paper analyzes the text language description characteristics of landslide geological hazards based on the mechanism of landslide geological hazards, formulates an annotation system and annotation specifications for semantic information of landslide geological hazards, and constructs a corpus for the field of landslide geological hazards. Meanwhile, the entity recognition experiments based on the corpus show that the accuracy, recall and precision of the named entity recognition model reach more than 90%, which verifies the applicability of the corpus and provides strong data support for the subsequent research work of landslide geology knowledge mapping.
Extracting valuable landslide geohazard entities from massive texts describing landslide geological hazards is the basis for constructing a landslide geohazard knowledge graph. Based on relevant unstructured text data such as landslide geological hazard exploration report documents, this paper analyzes the text language description characteristics of landslide geological hazards based on the mechanism of landslide geological hazards, formulates an annotation system and annotation specifications for semantic information of landslide geological hazards, and constructs a corpus for the field of landslide geological hazards. Meanwhile, the entity recognition experiments based on the corpus show that the accuracy, recall and precision of the named entity recognition model reach more than 90%, which verifies the applicability of the corpus and provides strong data support for the subsequent research work of landslide geology knowledge mapping.
Abstract:
The precision of path tracking is fundamental for the safe autonomous operation of intelligent vehicles. In order to address the vibration issue in sliding mode control systems and enhance the control precision of the path tracking controller, a novel Integral Sliding Mode Control strategy with Activation Function (PIDSM-AF) is proposed. Firstly, based on the two-degree-of-freedom vehicle model, the dynamic model of the vehicle is decomposed into a lateral error model to establish the lateral control model. Then, employing the extremum method, an integral sliding mode surface incorporating both heading angle error and lateral error is constructed. Considering the difficulty of eliminating system vibrations with general exponential convergence rates, this study introduces a nonlinear activation function based on the improved exponential convergence rate and designs a lateral path tracking controller based on the activation function sliding mode control. Finally, the improved sliding mode controller is subjected to double lane-change tests through Carsim-Simulink co-simulation. The results indicate that compared to the traditional terminal sliding mode controller, the tracking accuracy of the optimized integral sliding mode controller is improved by approximately 64% and 34.9% in maximum lateral position deviation under low-speed low-adhesion and high-speed high-adhesion conditions respectively, and the average lateral position deviation is improved by approximately 68.4% and 59.7%. Moreover, the optimized controller effectively suppresses the oscillations and overshoot variations of the vehicle"s heading angle and front wheel angle, demonstrating strong robustness.
The precision of path tracking is fundamental for the safe autonomous operation of intelligent vehicles. In order to address the vibration issue in sliding mode control systems and enhance the control precision of the path tracking controller, a novel Integral Sliding Mode Control strategy with Activation Function (PIDSM-AF) is proposed. Firstly, based on the two-degree-of-freedom vehicle model, the dynamic model of the vehicle is decomposed into a lateral error model to establish the lateral control model. Then, employing the extremum method, an integral sliding mode surface incorporating both heading angle error and lateral error is constructed. Considering the difficulty of eliminating system vibrations with general exponential convergence rates, this study introduces a nonlinear activation function based on the improved exponential convergence rate and designs a lateral path tracking controller based on the activation function sliding mode control. Finally, the improved sliding mode controller is subjected to double lane-change tests through Carsim-Simulink co-simulation. The results indicate that compared to the traditional terminal sliding mode controller, the tracking accuracy of the optimized integral sliding mode controller is improved by approximately 64% and 34.9% in maximum lateral position deviation under low-speed low-adhesion and high-speed high-adhesion conditions respectively, and the average lateral position deviation is improved by approximately 68.4% and 59.7%. Moreover, the optimized controller effectively suppresses the oscillations and overshoot variations of the vehicle"s heading angle and front wheel angle, demonstrating strong robustness.
Abstract:
Person re-identification (ReID) is a key technology in the field of intelligent video surveillance, aimed at retrieving the same person across cameras. Due to the complexity of monitoring scenarios, traditional single modal person re-identification is not suitable for some extreme situations such as low light and foggy days. In recent years, due to the needs of practical applications and the rapid development of deep learning, multi-modal person re-identification based on deep learning has received widespread attention. This article reviews the development of multi-modal person re-identification based on deep learning in recent years, elaborates on the shortcomings of traditional single modal person re-identification and summarizes the common application scenarios and advantages of multi-modal person re-identification, as well as the composition of each dataset. It focuses on analyzing the relevant methods and classifications of multi-modal person re-identification and discusses the hot spots and challenges of current research. Finally, the future development trends and potential application values of multi-modal person re-identification are prospected.
Person re-identification (ReID) is a key technology in the field of intelligent video surveillance, aimed at retrieving the same person across cameras. Due to the complexity of monitoring scenarios, traditional single modal person re-identification is not suitable for some extreme situations such as low light and foggy days. In recent years, due to the needs of practical applications and the rapid development of deep learning, multi-modal person re-identification based on deep learning has received widespread attention. This article reviews the development of multi-modal person re-identification based on deep learning in recent years, elaborates on the shortcomings of traditional single modal person re-identification and summarizes the common application scenarios and advantages of multi-modal person re-identification, as well as the composition of each dataset. It focuses on analyzing the relevant methods and classifications of multi-modal person re-identification and discusses the hot spots and challenges of current research. Finally, the future development trends and potential application values of multi-modal person re-identification are prospected.
Abstract:
Deep Neural Networks (DNNs) are vulnerable to specially designed adversarial examples and can be easily deceived. Although current detection techniques can identify some malicious inputs, their protective capabilities are still insufficient when facing complex attacks. This paper proposes a novel unsupervised adversarial example detection method based on unlabeled data. The core idea is to transform the adversarial example detection problem into an anomaly detection problem through feature construction and fusion. To this end, five core components are designed, including image transformation, neural network classifier, heatmap generation, distance calculation, and anomaly detector. Firstly, the original image is transformed, and the images before and after the transformation are input into the neural network classifier. The prediction probability array and convolutional layer features are extracted to generate a heatmap. The detector is extended from focusing solely on the model's output layer to the input layer features, enhancing the detector's ability to model and measure the differences between adversarial and normal samples. Then, the KL divergence of the probability arrays and the change distance of the heatmap focus points of the images before and after the transformation are calculated, and the distance features are input into the anomaly detector to determine whether it is an adversarial example. Experiments on the large-scale, high-quality image dataset ImageNet showed that our detector achieved an average AUC value of 0.77 against five different types of attacks, demonstrating good detection performance. Compared with other cutting-edge unsupervised adversarial example detectors, our detector has a TPR at least 15.88% higher while maintaining a similar false alarm rate, indicating a significant advantage in detection capability.
Deep Neural Networks (DNNs) are vulnerable to specially designed adversarial examples and can be easily deceived. Although current detection techniques can identify some malicious inputs, their protective capabilities are still insufficient when facing complex attacks. This paper proposes a novel unsupervised adversarial example detection method based on unlabeled data. The core idea is to transform the adversarial example detection problem into an anomaly detection problem through feature construction and fusion. To this end, five core components are designed, including image transformation, neural network classifier, heatmap generation, distance calculation, and anomaly detector. Firstly, the original image is transformed, and the images before and after the transformation are input into the neural network classifier. The prediction probability array and convolutional layer features are extracted to generate a heatmap. The detector is extended from focusing solely on the model's output layer to the input layer features, enhancing the detector's ability to model and measure the differences between adversarial and normal samples. Then, the KL divergence of the probability arrays and the change distance of the heatmap focus points of the images before and after the transformation are calculated, and the distance features are input into the anomaly detector to determine whether it is an adversarial example. Experiments on the large-scale, high-quality image dataset ImageNet showed that our detector achieved an average AUC value of 0.77 against five different types of attacks, demonstrating good detection performance. Compared with other cutting-edge unsupervised adversarial example detectors, our detector has a TPR at least 15.88% higher while maintaining a similar false alarm rate, indicating a significant advantage in detection capability.
Abstract:
The new power system needs the cooperative optimization scheduling of "integration of source-network-load-storage". At present, the scheduling automation system uses relational database to query and store data based on multiple associated tables, which is diffi-cult to meet the requirement of fast computation. It proposes a graph-based calculation method for day-day scheduling of load-network-load-storage in this paper. Firstly, the graph database is used to integrate the spatiotemporal data of source network. Second-ly, considering thermal power units, adjustable load, energy storage and other resources, it constructs the optimization model of in-tra-day scheduling of source-network-load-storage collaboration. Thirdly, based on the safety check results, the system operating state is corrected until all operating constraints are met. A quick calculation method of power flow based on graph calculation is proposed to check system security quickly. Finally, Through the analysis of improved IEEE118 nodes and IEEE1354 examples, it is verified that the proposed source network co-optimization can im-prove the computing efficiency.
The new power system needs the cooperative optimization scheduling of "integration of source-network-load-storage". At present, the scheduling automation system uses relational database to query and store data based on multiple associated tables, which is diffi-cult to meet the requirement of fast computation. It proposes a graph-based calculation method for day-day scheduling of load-network-load-storage in this paper. Firstly, the graph database is used to integrate the spatiotemporal data of source network. Second-ly, considering thermal power units, adjustable load, energy storage and other resources, it constructs the optimization model of in-tra-day scheduling of source-network-load-storage collaboration. Thirdly, based on the safety check results, the system operating state is corrected until all operating constraints are met. A quick calculation method of power flow based on graph calculation is proposed to check system security quickly. Finally, Through the analysis of improved IEEE118 nodes and IEEE1354 examples, it is verified that the proposed source network co-optimization can im-prove the computing efficiency.
Abstract:
The Yangtze River Delta (hereinafter referred to as the YRD), as one of the regions rich in forest resources in China, possesses immense carbon sequestration potential. Accurately capturing and assessing forest disturbances in the YRD is of great significance for enhancing forest management and ecological environmental protection in the region. This study employed the LandTrendr change detection algorithm and the method of Landsat 5-8 time-series stacking based on the Google Earth Engine (GEE) cloud platform to monitor forest disturbances and gains in the YRD over a span of 30 years from 1991 to 2020. The results indicated that the overall accuracy of the LandTrendr algorithm in monitoring forest disturbances and gains reached 88.5%, with both the producer's accuracy and user's accuracy exceeding 80%, demonstrating effective monitoring of forest disturbances and gains in the YRD. The total area of forest disturbances in the YRD was 9646.96 km2, while the total area of forest gains was 37205.46 km2. Over the past 30 years, the total forest area in the YRD has shown an increasing trend. The LandTrendr algorithm, based on the GEE cloud platform, has achieved precise monitoring of forest disturbances and gains in the YRD.
The Yangtze River Delta (hereinafter referred to as the YRD), as one of the regions rich in forest resources in China, possesses immense carbon sequestration potential. Accurately capturing and assessing forest disturbances in the YRD is of great significance for enhancing forest management and ecological environmental protection in the region. This study employed the LandTrendr change detection algorithm and the method of Landsat 5-8 time-series stacking based on the Google Earth Engine (GEE) cloud platform to monitor forest disturbances and gains in the YRD over a span of 30 years from 1991 to 2020. The results indicated that the overall accuracy of the LandTrendr algorithm in monitoring forest disturbances and gains reached 88.5%, with both the producer's accuracy and user's accuracy exceeding 80%, demonstrating effective monitoring of forest disturbances and gains in the YRD. The total area of forest disturbances in the YRD was 9646.96 km2, while the total area of forest gains was 37205.46 km2. Over the past 30 years, the total forest area in the YRD has shown an increasing trend. The LandTrendr algorithm, based on the GEE cloud platform, has achieved precise monitoring of forest disturbances and gains in the YRD.
Abstract:
Aiming at the problems of multi-sensor information fusion, a UAV position estimation based on non Gaussian strength and fused CKF is proposed. Firstly, to address the issue of poor estimation and compensation of system errors, a method of constructing pseudo measurement and equivalent measurement equations for radar system errors is proposed, and Kalman filtering is used for real-time estimation of radar system errors. Secondly, to address the issue of sample distribution morphology not being taken into account in skewness kurtosis testing, a method based on multimodal distribution and skewness coeffi- -cient constructors is proposed. By using multimodal distribution to identify multimodal distribution features and calculate sample skewness, non Gaussian features of data can be more comprehensively evaluated. Thirdly, to solve the problem of difficulty in constructing fusion coefficients due to non Gaussian noise in multi-sensor fusion, a method of constructing sensor fusion coefficients based on the strength of non Gaussian noise is proposed. Finally, using Cauchy kernel and Gaussian kernel CKF, the problem of reduced estimation accuracy caused by non Gaussian noise in dynamic state estimation is solved.
Aiming at the problems of multi-sensor information fusion, a UAV position estimation based on non Gaussian strength and fused CKF is proposed. Firstly, to address the issue of poor estimation and compensation of system errors, a method of constructing pseudo measurement and equivalent measurement equations for radar system errors is proposed, and Kalman filtering is used for real-time estimation of radar system errors. Secondly, to address the issue of sample distribution morphology not being taken into account in skewness kurtosis testing, a method based on multimodal distribution and skewness coeffi- -cient constructors is proposed. By using multimodal distribution to identify multimodal distribution features and calculate sample skewness, non Gaussian features of data can be more comprehensively evaluated. Thirdly, to solve the problem of difficulty in constructing fusion coefficients due to non Gaussian noise in multi-sensor fusion, a method of constructing sensor fusion coefficients based on the strength of non Gaussian noise is proposed. Finally, using Cauchy kernel and Gaussian kernel CKF, the problem of reduced estimation accuracy caused by non Gaussian noise in dynamic state estimation is solved.
Abstract:
Aiming at the problems that the rotor speed and position are difficult to be accurately estimated by the sliding mode observer (SMO) due to high frequency chattering, and the robustness and speed tracking performance of the speed loop PI controller are poor, an improved SMO-based speed sensorless control for permanent magnet synchronous motor(PMSM) is proposed. First, an infinite impulse response(IIR) filter is added to the sliding mode observer in the direct torque control(DTC) system to deal with the high frequency noise in the output of the sliding mode observer, Then a continuous hyperbolic tangent function is adopted to replace the discontinuous sign function to attenuate the system chattering. Moreover, an active disturbance rejection control(ADRC)-based speed loop controller is to improve the confliction between the overshoot and settling time of the system. Finally, a hardware-in-the-loop experimental platform is built in Matlab/Simulink environment. The simulation and experimental results show that under the same load and interference conditions, compared with the traditional speed sensorless control-system, the proposed speed sensorless ADRC-based controller can suppress the chattering effectively and exhibits better robustness and speed tracking performance.
Aiming at the problems that the rotor speed and position are difficult to be accurately estimated by the sliding mode observer (SMO) due to high frequency chattering, and the robustness and speed tracking performance of the speed loop PI controller are poor, an improved SMO-based speed sensorless control for permanent magnet synchronous motor(PMSM) is proposed. First, an infinite impulse response(IIR) filter is added to the sliding mode observer in the direct torque control(DTC) system to deal with the high frequency noise in the output of the sliding mode observer, Then a continuous hyperbolic tangent function is adopted to replace the discontinuous sign function to attenuate the system chattering. Moreover, an active disturbance rejection control(ADRC)-based speed loop controller is to improve the confliction between the overshoot and settling time of the system. Finally, a hardware-in-the-loop experimental platform is built in Matlab/Simulink environment. The simulation and experimental results show that under the same load and interference conditions, compared with the traditional speed sensorless control-system, the proposed speed sensorless ADRC-based controller can suppress the chattering effectively and exhibits better robustness and speed tracking performance.
Abstract:
This paper proposes a data-driven regionalization algorithm based on machine learning and modern statistical methods to address current issues such as limited climate regionalization variables, underutilized information, and insufficient consideration of climate change impacts. Firstly, we use the Mann-Kendall test and sliding t-test to identify change points of time series of primary variables and segment the study period accordingly. Next, we employ BP canonical correlation analysis to select covariates and establish a multivariate Self-Organizing Map (SOM) clustering algorithm to achieve climate regionalization for different stages. Finally, we analyze the practical significance of regionalization results in combination with climate zone profiles, and assess the impact of climate change on climate regionalization. Experimental results demonstrate that the proposed regionalization algorithm, driven by data rather than contour lines of primary variables or manually set thresholds, improves data utilization and ensures a more objective and rational regionalization process. By incorporating multiple covariates and climate change impacts into the algorithm, the efficiency and reliability of regionalization are effectively enhanced without considering climate background during the regionalization process.
This paper proposes a data-driven regionalization algorithm based on machine learning and modern statistical methods to address current issues such as limited climate regionalization variables, underutilized information, and insufficient consideration of climate change impacts. Firstly, we use the Mann-Kendall test and sliding t-test to identify change points of time series of primary variables and segment the study period accordingly. Next, we employ BP canonical correlation analysis to select covariates and establish a multivariate Self-Organizing Map (SOM) clustering algorithm to achieve climate regionalization for different stages. Finally, we analyze the practical significance of regionalization results in combination with climate zone profiles, and assess the impact of climate change on climate regionalization. Experimental results demonstrate that the proposed regionalization algorithm, driven by data rather than contour lines of primary variables or manually set thresholds, improves data utilization and ensures a more objective and rational regionalization process. By incorporating multiple covariates and climate change impacts into the algorithm, the efficiency and reliability of regionalization are effectively enhanced without considering climate background during the regionalization process.
Abstract:
Aiming at the problems of strong fluctuation of wind power and low prediction accuracy, An improved logistic-t-dung beetle optimizer (LTDBO) is proposed to optimize variational mode decomposition. VMD) parameters and LTDBO algorithm optimize the hybrid short-term wind power prediction model of long short-term memory network (LSTM) hyperparameters. First, the kurtotic of the mean envelope spectrum is used as the fitness function, and LTDBO algorithm is used to optimize the decomposition layers and penalty factors of VMD. Then, the wind power sequence after data cleaning is decomposed with VMD to obtain the stationary intrinsic mode function (IMF) of different frequencies. Each IMF input was predicted by LSTM with LTDBO's hyperparameter optimization. Finally, each IMF predicted value was superposed and reconstructed to obtain the final result. The experimental results show that the LTDBO algorithm can find the optimal combination of VMD and LSTM hyperparameters, and the combined model of LTDBO-VMD-LTDBO-LSTM has better prediction accuracy and robustness in the field of wind power prediction.
Aiming at the problems of strong fluctuation of wind power and low prediction accuracy, An improved logistic-t-dung beetle optimizer (LTDBO) is proposed to optimize variational mode decomposition. VMD) parameters and LTDBO algorithm optimize the hybrid short-term wind power prediction model of long short-term memory network (LSTM) hyperparameters. First, the kurtotic of the mean envelope spectrum is used as the fitness function, and LTDBO algorithm is used to optimize the decomposition layers and penalty factors of VMD. Then, the wind power sequence after data cleaning is decomposed with VMD to obtain the stationary intrinsic mode function (IMF) of different frequencies. Each IMF input was predicted by LSTM with LTDBO's hyperparameter optimization. Finally, each IMF predicted value was superposed and reconstructed to obtain the final result. The experimental results show that the LTDBO algorithm can find the optimal combination of VMD and LSTM hyperparameters, and the combined model of LTDBO-VMD-LTDBO-LSTM has better prediction accuracy and robustness in the field of wind power prediction.
Abstract:
There are various remote sensing data sources for biomass estimation, including optical, synthetic aperture radar (SAR) and light detection and ranging (LiDAR). When using remote sensing techniques for regional biomass estimation, the scarcity of field plot data constrains the improvement of model inversion accuracy. In this study, unmanned aerial vehicle LiDAR point cloud data combined with field survey was used to expand the sample size, trying to verify the feasibility of using airborne radar instead of field survey, and establishing an inversion model of vegetation aboveground biomass with optical images features(Vegetation index data and texture feature data). The results show that: 1) The biomass estimated based on the tree crown width data obtained from LiDAR has a high level of accuracy, with an overall error percentage of 1.74% compared to the biomass results obtained from field measurements.; 2) the aboveground biomass density estimated from the LiDAR data is 96.78 Mg·hm-2, the biomass density from the optical image feature model is 107.94 Mg·hm-2 in the Longwang Hill; 3) The validation data from the campus adjacent to Longwang Mountain shows that the model-inverted aboveground biomass density is 92.6 Mg·hm-2, while the result based on LiDAR data is 104.11 Mg·hm-2. It can be observed that this method demonstrates good effectiveness. Therefore, the sample size can be expanded by unmanned aerial vehicle LiDAR and crown amplitude biomass model, and the resulting image characteristic biomass model has good results and provides a feasible method for large-scale biomass inversion.
There are various remote sensing data sources for biomass estimation, including optical, synthetic aperture radar (SAR) and light detection and ranging (LiDAR). When using remote sensing techniques for regional biomass estimation, the scarcity of field plot data constrains the improvement of model inversion accuracy. In this study, unmanned aerial vehicle LiDAR point cloud data combined with field survey was used to expand the sample size, trying to verify the feasibility of using airborne radar instead of field survey, and establishing an inversion model of vegetation aboveground biomass with optical images features(Vegetation index data and texture feature data). The results show that: 1) The biomass estimated based on the tree crown width data obtained from LiDAR has a high level of accuracy, with an overall error percentage of 1.74% compared to the biomass results obtained from field measurements.; 2) the aboveground biomass density estimated from the LiDAR data is 96.78 Mg·hm-2, the biomass density from the optical image feature model is 107.94 Mg·hm-2 in the Longwang Hill; 3) The validation data from the campus adjacent to Longwang Mountain shows that the model-inverted aboveground biomass density is 92.6 Mg·hm-2, while the result based on LiDAR data is 104.11 Mg·hm-2. It can be observed that this method demonstrates good effectiveness. Therefore, the sample size can be expanded by unmanned aerial vehicle LiDAR and crown amplitude biomass model, and the resulting image characteristic biomass model has good results and provides a feasible method for large-scale biomass inversion.
Abstract:
In view of the existing image inpainting algorithms, the texture may be blurred and the instability in the training process of the repaired image, a generative adversarial network image inpainting algorithm based on the diffusion process was proposed. The diffusion model is introduced into the dual discriminator generation adversarial network, and the image generated by the generator and the real image go through the forward diffusion process to obtain the inverted image with Gaussian noise and the real image, which is used as the input of the discriminator, which improves the repair quality and increases the stability of model training. Style loss and perceptual loss are introduced into the loss function to learn semantic feature differences, eliminate motion blur, and retain more details and edge information in the repair results. Qualitative and quantitative analysis and ablation experiments were done on the datasets CelebA and Places2, respectively, and according to the evaluation results and restoration effects show that the proposed algorithms have better performance. Compared with the current repair method, peak signal-to-noise ratio and structural similarity were improved by an average of 1.26db and 1.84%, respectively, and the L1 error was decreased by an average of 25.7%, and according to the change of the loss function it is seen that after the The training of image restoration algorithm with diffusion process is more stable.
In view of the existing image inpainting algorithms, the texture may be blurred and the instability in the training process of the repaired image, a generative adversarial network image inpainting algorithm based on the diffusion process was proposed. The diffusion model is introduced into the dual discriminator generation adversarial network, and the image generated by the generator and the real image go through the forward diffusion process to obtain the inverted image with Gaussian noise and the real image, which is used as the input of the discriminator, which improves the repair quality and increases the stability of model training. Style loss and perceptual loss are introduced into the loss function to learn semantic feature differences, eliminate motion blur, and retain more details and edge information in the repair results. Qualitative and quantitative analysis and ablation experiments were done on the datasets CelebA and Places2, respectively, and according to the evaluation results and restoration effects show that the proposed algorithms have better performance. Compared with the current repair method, peak signal-to-noise ratio and structural similarity were improved by an average of 1.26db and 1.84%, respectively, and the L1 error was decreased by an average of 25.7%, and according to the change of the loss function it is seen that after the The training of image restoration algorithm with diffusion process is more stable.
Abstract:
In order to guarantee the reliable and economical operation of the standalone micro grid, the optimal power capacity configuration need to be carried in the power system planning. The power capacity of independent micro grid is not only affected by regional load but also by natural resources such as the wind and solar energy. In this paper, the characteristics of natural resource condition in different regions are analyzed. A multiple objective optimization model with the goal of minimizing the annual generation cost is established, considering the reliability of power supply and electricity environmental protection. Constraints of electric power and energy balance are taken into account. The linearization algorithm is applied to solve the optimization model. The output scenario of renewable energy is simulated by the generative adversarial networks (GAN), and reduced by the improved K-Medoids clustering algorithm. In order to analyze the characteristics of the natural resources, the index evaluation system is established, and the evaluation grade of natural resources in 32 provinces of China is obtained based on the fuzzy evaluation method. This paper compares the power capacity configuration of standalone microgrid in five typical regions with different natural resources, hoping to provide some reference for the planning of standalone microgrid system.
In order to guarantee the reliable and economical operation of the standalone micro grid, the optimal power capacity configuration need to be carried in the power system planning. The power capacity of independent micro grid is not only affected by regional load but also by natural resources such as the wind and solar energy. In this paper, the characteristics of natural resource condition in different regions are analyzed. A multiple objective optimization model with the goal of minimizing the annual generation cost is established, considering the reliability of power supply and electricity environmental protection. Constraints of electric power and energy balance are taken into account. The linearization algorithm is applied to solve the optimization model. The output scenario of renewable energy is simulated by the generative adversarial networks (GAN), and reduced by the improved K-Medoids clustering algorithm. In order to analyze the characteristics of the natural resources, the index evaluation system is established, and the evaluation grade of natural resources in 32 provinces of China is obtained based on the fuzzy evaluation method. This paper compares the power capacity configuration of standalone microgrid in five typical regions with different natural resources, hoping to provide some reference for the planning of standalone microgrid system.
Abstract:
In addressing challenges such as small target sizes, blurred target features, and difficulty distinguishing between targets and backgrounds in small object detection tasks, a method based on dual-stream contrastive feature learning and image multi-scale degradation augmentation is proposed. Firstly, the input images of the contrastive learning model are subjected to multi-scale degradation augmentation, enhancing the model's perception of capturing small targets. Secondly, contrastive learning representations are conducted in both spatial and frequency domains simultaneously to learn more discriminative target recognition features, improving the model's ability to differentiate between targets and backgrounds and thus enhancing small object detection. To validate the effectiveness of the proposed method, ablation experiments are designed, and the detection performance is compared with other advanced algorithms. Experimental results show that the proposed method achieves a 3.6% improvement in average precision (mAP) compared to the baseline algorithm on the MS COCO datasets, a 7.7% improvement in average precision for small objects (APS) compared to mainstream advanced algorithms. On the VisDrone2019 datasets, the proposed method achieves a 2.4% increase in mAP compared to the baseline algorithm, demonstrating comprehensive performance superiority over the baseline algorithm and other mainstream advanced algorithms. Visual analysis of detection results indicates significant improvements in false negatives and false positives for small object detection using the proposed method.
In addressing challenges such as small target sizes, blurred target features, and difficulty distinguishing between targets and backgrounds in small object detection tasks, a method based on dual-stream contrastive feature learning and image multi-scale degradation augmentation is proposed. Firstly, the input images of the contrastive learning model are subjected to multi-scale degradation augmentation, enhancing the model's perception of capturing small targets. Secondly, contrastive learning representations are conducted in both spatial and frequency domains simultaneously to learn more discriminative target recognition features, improving the model's ability to differentiate between targets and backgrounds and thus enhancing small object detection. To validate the effectiveness of the proposed method, ablation experiments are designed, and the detection performance is compared with other advanced algorithms. Experimental results show that the proposed method achieves a 3.6% improvement in average precision (mAP) compared to the baseline algorithm on the MS COCO datasets, a 7.7% improvement in average precision for small objects (APS) compared to mainstream advanced algorithms. On the VisDrone2019 datasets, the proposed method achieves a 2.4% increase in mAP compared to the baseline algorithm, demonstrating comprehensive performance superiority over the baseline algorithm and other mainstream advanced algorithms. Visual analysis of detection results indicates significant improvements in false negatives and false positives for small object detection using the proposed method.
Abstract:
Path tracking is essential for unmanned driving. This study presents the design of a path tracking system for unmanned trucks, aiming to enhance accuracy and stability across various speeds. The system employs a linear quadratic regulator (LQR) optimized through an improved genetic algorithm. Firstly, this study begins by establishing a two-degree-of-freedom dynamic model and tracking error model of the vehicle based on the natural coordinate system. Subsequently, a LQR controller is designed to eliminate steady-state errors and enhance tracking accuracy through the implementation of feedforward control. Secondly, the genetic algorithm is enhanced to optimize the weight matrix of the LQR controller, resulting in improved accuracy and stability for path tracking. Finally, the control effectiveness of the designed LQR controller was simulated and verified across a range of operating conditions using the joint simulation platform of Matlab/Simulink and TruckSim. The findings demonstrate that the tracking accuracy of the LQR controller, optimized using GA, has exhibited an average improvement of approximately 38.8%. Additionally, it has demonstrated enhanced stability. Specifically, the position error and heading error could be maintained within 0.17m and 0.11rad, respectively, thereby validating the efficacy of the tracking control framework proposed in this research paper.
Path tracking is essential for unmanned driving. This study presents the design of a path tracking system for unmanned trucks, aiming to enhance accuracy and stability across various speeds. The system employs a linear quadratic regulator (LQR) optimized through an improved genetic algorithm. Firstly, this study begins by establishing a two-degree-of-freedom dynamic model and tracking error model of the vehicle based on the natural coordinate system. Subsequently, a LQR controller is designed to eliminate steady-state errors and enhance tracking accuracy through the implementation of feedforward control. Secondly, the genetic algorithm is enhanced to optimize the weight matrix of the LQR controller, resulting in improved accuracy and stability for path tracking. Finally, the control effectiveness of the designed LQR controller was simulated and verified across a range of operating conditions using the joint simulation platform of Matlab/Simulink and TruckSim. The findings demonstrate that the tracking accuracy of the LQR controller, optimized using GA, has exhibited an average improvement of approximately 38.8%. Additionally, it has demonstrated enhanced stability. Specifically, the position error and heading error could be maintained within 0.17m and 0.11rad, respectively, thereby validating the efficacy of the tracking control framework proposed in this research paper.
Abstract:
In the task of power equipment fault detection, the performance of the model is affected by various factors, such as the diversity of fault types, the complexity of fault characteristics, and the differences in image quality. To solve these problems, a new power equipment fault detection model based on TrellisNet and attention mechanism is proposed. The model first integrates Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) to construct LSTM-CNN to obtain fault characteristics in images, which can effectively distinguish features of different fault types and reduce the influence of noise and interference factors. In addition, the feature data obtained by LSTM-CNN is used as input, and by embedding the attention mechanism into TrellisNet, an AT-TrellisNet network with high resolution is constructed to detect the fault types of different power equipment. Finally, by selecting five common power equipment faults for model verification, the experimental results show that the model has a higher detection accuracy rate of over 90% compared to some existing detection models, which can meet the actual power equipment fault detection requirements.
In the task of power equipment fault detection, the performance of the model is affected by various factors, such as the diversity of fault types, the complexity of fault characteristics, and the differences in image quality. To solve these problems, a new power equipment fault detection model based on TrellisNet and attention mechanism is proposed. The model first integrates Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) to construct LSTM-CNN to obtain fault characteristics in images, which can effectively distinguish features of different fault types and reduce the influence of noise and interference factors. In addition, the feature data obtained by LSTM-CNN is used as input, and by embedding the attention mechanism into TrellisNet, an AT-TrellisNet network with high resolution is constructed to detect the fault types of different power equipment. Finally, by selecting five common power equipment faults for model verification, the experimental results show that the model has a higher detection accuracy rate of over 90% compared to some existing detection models, which can meet the actual power equipment fault detection requirements.
Abstract:
Aiming at the complexity and nonlinearity of natural gas load sequence, this paper proposes a natural gas load combination forecasting model based on Time2Vec. Firstly, Time2vec, a time vector embedding layer, is introduced to convert the time series into a continuous vector space, which improves the computational efficiency of the model on the time series information; secondly, the Pearson correlation coefficient is used for the correlation analysis, and the meteorological features with strong correlation are extracted as inputs. The temporal features extracted by Time2Vec and the meteorological features selected by Pearson"s correlation coefficient are inputted into the two models of Long Short-Term Memory Network (LSTM) and Temporal Convolutional Network (TCN) for prediction, which make full use of the long term memory capability of LSTM and the local feature extraction capability of TCN. Finally, these two models are combined through Attention mechanism (Attention), and different weights are assigned according to the importance of the two to obtain the final prediction results. The experimental results show that the proposed Time2Vec-based natural gas load combination prediction model has stronger adaptability and higher accuracy.
Aiming at the complexity and nonlinearity of natural gas load sequence, this paper proposes a natural gas load combination forecasting model based on Time2Vec. Firstly, Time2vec, a time vector embedding layer, is introduced to convert the time series into a continuous vector space, which improves the computational efficiency of the model on the time series information; secondly, the Pearson correlation coefficient is used for the correlation analysis, and the meteorological features with strong correlation are extracted as inputs. The temporal features extracted by Time2Vec and the meteorological features selected by Pearson"s correlation coefficient are inputted into the two models of Long Short-Term Memory Network (LSTM) and Temporal Convolutional Network (TCN) for prediction, which make full use of the long term memory capability of LSTM and the local feature extraction capability of TCN. Finally, these two models are combined through Attention mechanism (Attention), and different weights are assigned according to the importance of the two to obtain the final prediction results. The experimental results show that the proposed Time2Vec-based natural gas load combination prediction model has stronger adaptability and higher accuracy.
Abstract:
In static grid maps, an improved algorithm that combines Jump Point Search and Bi-directional Parallel Ant Colony Search is proposed to address the problem of slow convergence and easy trapping in local optima of traditional ant colony algorithms for AGV path planning. Firstly, the actual research environment is modeled by gridization, and the improved Jump Point Search algorithm is used to generate the initial suboptimal path for bi-directional search, providing a reference for the initial search direction of bi-directional ant colony search. Secondly, an improved transition probability heuristic function is used in the bi-directional parallel ant colony search process, which considers the avoidance of collision between AGV and obstacles when determining the next transition node. Meanwhile, by designing an information sharing mechanism and combining two fusion models of improved information increment and concentration, the global information concentration is shared and updated to better explore and optimize the path and ensure the connection of bi-directional paths. Finally, experimental results are compared with traditional ant colony algorithms to verify the improved algorithm"s global search capability, efficiency and safety.
In static grid maps, an improved algorithm that combines Jump Point Search and Bi-directional Parallel Ant Colony Search is proposed to address the problem of slow convergence and easy trapping in local optima of traditional ant colony algorithms for AGV path planning. Firstly, the actual research environment is modeled by gridization, and the improved Jump Point Search algorithm is used to generate the initial suboptimal path for bi-directional search, providing a reference for the initial search direction of bi-directional ant colony search. Secondly, an improved transition probability heuristic function is used in the bi-directional parallel ant colony search process, which considers the avoidance of collision between AGV and obstacles when determining the next transition node. Meanwhile, by designing an information sharing mechanism and combining two fusion models of improved information increment and concentration, the global information concentration is shared and updated to better explore and optimize the path and ensure the connection of bi-directional paths. Finally, experimental results are compared with traditional ant colony algorithms to verify the improved algorithm"s global search capability, efficiency and safety.
Abstract:
To solve the halo effect in edge area and color distortion problems in sky area of the classic image dehazing algorithm, a new improved dark channel prior dehazing method based on sky detection and super-pixel segmentation is proposed. The proposed method first applied the HSV color transformation to extract the light component of haze image with an adaptive-threshold segmentation technology. Then sky and non-sky areas were extracted with image connectivity analysis technology. After that, the atmospheric light value was estimated from sky areas, and transmission map of sky and non-sky areas were estimated with luminance model and super-pixels dark channel prior model separately, and a pixel-based fusion model was proposed to obtain a comprehensive transmission map to promote the smooth transition in boundary area. Next, the transmission map was refined by a multi-scale guided filter algorithms. Finally, the dehazed image was restored naturally with atmospheric scattering model and brightness enhancement processing. Experimental results show that the sky area identified from this method is more continuously and completely, and it is effectively to overcome the influence of halo effects when use superpixels instead of square windows to acquire transmission maps. The estimation of atmospheric light values and transmittance maps are more objectively and accurately. From the perspectives of subjective qualitative and objective quantitative evaluation, the proposed method shows more advantage over small overall error, excellent signal-to-noise ratio, and high structural similarity in dehazed images. The proposed method can restore a more natural sky and weaken the halo effect in edge area, the quality of the dehazed image is further improved compared to the state of the art on both qualitative and quantitative analysis.
To solve the halo effect in edge area and color distortion problems in sky area of the classic image dehazing algorithm, a new improved dark channel prior dehazing method based on sky detection and super-pixel segmentation is proposed. The proposed method first applied the HSV color transformation to extract the light component of haze image with an adaptive-threshold segmentation technology. Then sky and non-sky areas were extracted with image connectivity analysis technology. After that, the atmospheric light value was estimated from sky areas, and transmission map of sky and non-sky areas were estimated with luminance model and super-pixels dark channel prior model separately, and a pixel-based fusion model was proposed to obtain a comprehensive transmission map to promote the smooth transition in boundary area. Next, the transmission map was refined by a multi-scale guided filter algorithms. Finally, the dehazed image was restored naturally with atmospheric scattering model and brightness enhancement processing. Experimental results show that the sky area identified from this method is more continuously and completely, and it is effectively to overcome the influence of halo effects when use superpixels instead of square windows to acquire transmission maps. The estimation of atmospheric light values and transmittance maps are more objectively and accurately. From the perspectives of subjective qualitative and objective quantitative evaluation, the proposed method shows more advantage over small overall error, excellent signal-to-noise ratio, and high structural similarity in dehazed images. The proposed method can restore a more natural sky and weaken the halo effect in edge area, the quality of the dehazed image is further improved compared to the state of the art on both qualitative and quantitative analysis.
Abstract:
To effectively solve the problem of low visual quality of stego-image in existing color image Reversible Data Hiding (RDH) algorithm, a novel RDH scheme based on multi-level interpolation prediction and global sorting is proposed. Firstly, to make full use of the features of different texture regions in the image, a multi-level interpolation prediction algorithm is designed to significantly improve the prediction accuracy of pixels. Then, a global sorting strategy based on complexity is designed to sort the prediction errors in the three channels of color images respectively, making full use of the global characteristics of the prediction errors in each channel. A more concentrated Three-Dimensional Prediction Error Histogram (3D PEH) is generated. Finally, the adaptive 3D mapping strategy is used to modify the error histogram and embed the secret data. Experimental results show that the proposed method achieves better embedding performance compared with the latest schemes.
To effectively solve the problem of low visual quality of stego-image in existing color image Reversible Data Hiding (RDH) algorithm, a novel RDH scheme based on multi-level interpolation prediction and global sorting is proposed. Firstly, to make full use of the features of different texture regions in the image, a multi-level interpolation prediction algorithm is designed to significantly improve the prediction accuracy of pixels. Then, a global sorting strategy based on complexity is designed to sort the prediction errors in the three channels of color images respectively, making full use of the global characteristics of the prediction errors in each channel. A more concentrated Three-Dimensional Prediction Error Histogram (3D PEH) is generated. Finally, the adaptive 3D mapping strategy is used to modify the error histogram and embed the secret data. Experimental results show that the proposed method achieves better embedding performance compared with the latest schemes.
Abstract:
Fault location plays an important role in long distance HVDC transmission system. To solve the problems of inaccurate attenuation coefficient calculation and difficult capture of secondary wave head, a fault location model based on improved Pelican optimization algorithm (IPOA) and optimized least squares support vector Machine (LSSVM) is proposed. Firstly, according to the attenuation principle of traveling wave, the formula of fault distance and modulus maximum ratio of two terminal mode components of the line is derived, and the nonlinear relationship between them is found. Secondly, using LSSVM to generalize the relationship between the two, the improved POA algorithm is used to optimize the key parameters of LSSVM, and the IPOA-LSSVM fault location model is established. By collecting fault signals at both ends, the amplitude ratio of the first wave head is obtained as the input of the model, and the fault distance is used as the output for simulation verification. The simulation results show that the model is not affected by the transition resistance and fault type, and can locate reliably and accurately.
Fault location plays an important role in long distance HVDC transmission system. To solve the problems of inaccurate attenuation coefficient calculation and difficult capture of secondary wave head, a fault location model based on improved Pelican optimization algorithm (IPOA) and optimized least squares support vector Machine (LSSVM) is proposed. Firstly, according to the attenuation principle of traveling wave, the formula of fault distance and modulus maximum ratio of two terminal mode components of the line is derived, and the nonlinear relationship between them is found. Secondly, using LSSVM to generalize the relationship between the two, the improved POA algorithm is used to optimize the key parameters of LSSVM, and the IPOA-LSSVM fault location model is established. By collecting fault signals at both ends, the amplitude ratio of the first wave head is obtained as the input of the model, and the fault distance is used as the output for simulation verification. The simulation results show that the model is not affected by the transition resistance and fault type, and can locate reliably and accurately.
Abstract:
Line losses are an inevitable issue in the process of power distribution and have a direct impact on the economic performance of power companies. Therefore, accurately predicting line losses has become a focal concern for the power sector. Predicting line losses using recurrent neural networks can encounter issues related to forgetting, making accurate predictions challenging for longer time series. To enhance the accuracy of line loss prediction, a line loss prediction model based on the self-attention mechanism is proposed. Initially, a high-dimensional time series is constructed using information such as the overall power consumption, line losses, and time within a recent period. Subsequently, a self-attention neural network model is established. Then, an open-source dataset is utilized for training and testing, with mean squared error as evaluation metrics. Ultimately, experimental results demonstrate that the line loss prediction model based on the self-attention mechanism can effectively predict line losses, particularly demonstrating strong predictive performance for longer line loss sequences when employing self-attention neural networks.
Line losses are an inevitable issue in the process of power distribution and have a direct impact on the economic performance of power companies. Therefore, accurately predicting line losses has become a focal concern for the power sector. Predicting line losses using recurrent neural networks can encounter issues related to forgetting, making accurate predictions challenging for longer time series. To enhance the accuracy of line loss prediction, a line loss prediction model based on the self-attention mechanism is proposed. Initially, a high-dimensional time series is constructed using information such as the overall power consumption, line losses, and time within a recent period. Subsequently, a self-attention neural network model is established. Then, an open-source dataset is utilized for training and testing, with mean squared error as evaluation metrics. Ultimately, experimental results demonstrate that the line loss prediction model based on the self-attention mechanism can effectively predict line losses, particularly demonstrating strong predictive performance for longer line loss sequences when employing self-attention neural networks.
Abstract:
Reliable and effective medium- and long-term power demand forecast is an important basis for power production and delivery, while at present, China's new energy planning is developing rapidly, the impact of wind and light volatility can not be ignored, the future of the power system planning can adapt to the demand change scenarios economically and efficiently run has become a high concern. In order to comprehensively consider the power demand and power system cooperative operation, a predictive dispatch-integrated evaluation model based on the bat algorithm optimization of the extreme learning machine algorithm and the introduction of fuzzy parameters of the source-load-storage resource cooperative operation algorithm is proposed, and an analysis and research is carried out for the example of the region in the Northwest, and the results show that the model can accurately forecast the power demand under different development scenarios, and can provide scientific references for the source-load-storage resource The results show that the model can accurately predict power demand under different development scenarios, and can provide scientific references for source-load-storage resource planning.
Reliable and effective medium- and long-term power demand forecast is an important basis for power production and delivery, while at present, China's new energy planning is developing rapidly, the impact of wind and light volatility can not be ignored, the future of the power system planning can adapt to the demand change scenarios economically and efficiently run has become a high concern. In order to comprehensively consider the power demand and power system cooperative operation, a predictive dispatch-integrated evaluation model based on the bat algorithm optimization of the extreme learning machine algorithm and the introduction of fuzzy parameters of the source-load-storage resource cooperative operation algorithm is proposed, and an analysis and research is carried out for the example of the region in the Northwest, and the results show that the model can accurately forecast the power demand under different development scenarios, and can provide scientific references for the source-load-storage resource The results show that the model can accurately predict power demand under different development scenarios, and can provide scientific references for source-load-storage resource planning.
Abstract:
In this paper, we propose a pruning and optimization algorithm (GWP) based on gradient weight Pursuit to solve the overfitting problem in the unsupervised domain, that is, the accuracy of downstream tasks is much lower than that of training sets. The dense-sparse-dense strategy is applied to solve the overfitting problem for the difference and adversarial adaptive methods in the unsupervised domain. The network is trained intensively and which connections are important are learned. The second is the pruning stage. Different from the pruning process in the original dense-sparse-dense strategy, the optimization algorithm in this paper considers both weight and gradient. On the one hand, weight information (i.e. zero-order information) is used, and on the other hand, the influence of gradient information (i.e. first-order information) on the network pruning process is also considered. In the final intensive phase, the pruned connections are restored and the intensive network is retrained with a smaller learning rate. Finally, the obtained network achieves ideal results in downstream tasks. The experimental results show that the proposed GWTA can effectively improve the accuracy of downstream tasks and has a plug-and-play effect compared with the original difference-based and adversarial domain adaptive methods.
In this paper, we propose a pruning and optimization algorithm (GWP) based on gradient weight Pursuit to solve the overfitting problem in the unsupervised domain, that is, the accuracy of downstream tasks is much lower than that of training sets. The dense-sparse-dense strategy is applied to solve the overfitting problem for the difference and adversarial adaptive methods in the unsupervised domain. The network is trained intensively and which connections are important are learned. The second is the pruning stage. Different from the pruning process in the original dense-sparse-dense strategy, the optimization algorithm in this paper considers both weight and gradient. On the one hand, weight information (i.e. zero-order information) is used, and on the other hand, the influence of gradient information (i.e. first-order information) on the network pruning process is also considered. In the final intensive phase, the pruned connections are restored and the intensive network is retrained with a smaller learning rate. Finally, the obtained network achieves ideal results in downstream tasks. The experimental results show that the proposed GWTA can effectively improve the accuracy of downstream tasks and has a plug-and-play effect compared with the original difference-based and adversarial domain adaptive methods.
Abstract:
The Aksu River Basin landscape engineering, as a flagship project of the UN for ecological restoration, plays an important role in addressing climate change, protecting biodiversity, enhancing the value of ecological services and the quality of ecological products, and maintaining ecological security in the basin. Based on the law of regional differentiation, the principle of ecosystem coupling, and the mechanism of environmental effects, and based on the analysis of the coupling relationship between the spatiotemporal evolution of ecological vulnerability and ecological security, a ecological fragility evaluation index system for the Aksu River Basin"s mountains, rivers, forests, fields, lakes, grasslands, and sands system is constructed, consisting of 4 subsystems and 20 indicators. And apply the EFI method to analyze the contribution of various indicators and subsystems to EFI and their ecological effects.The larger the positive indicator value of EFI, the lower the ecological security.The larger the absolute value of the negative EFI indicator, the higher the ecological security. Obtained multi-source attribute data for various indicators in 1998 and 2023 through practical monitoring, experimental analysis, model simulation, remote sensing mapping, logical reasoning and empirical analysis, the weights and thresholds of each indicator are determined,which provide a data basis for ecological security evaluation through normalization and other methods. The results indicate that: (1) Ecological security is a complex and systematic issue. The interaction between various subsystems and their elements shows a certain degree of comprehensiveness, lag, complexity, and uncertainty in the ecological security of the basin. (2) The ecological security of the Aksu River Basin has been in a stable and improving state in recent 25 years. The EFI values are 0.08 and 0.06 in 1998 and 2023 respectively. According to the evaluation criteria, ecological fragility is further weakening and the degree of ecological security is continuously increasing. (3) The ecological security level of each subsystem and its contribution to the ecological security effect of the basin are different from each other. The indicators of the water resources subsystem are basically stable and gradually improving, and the fragility of the subsystem is becoming weaker while maintaining stability. The ecological security characteristics show an evolution pattern of becoming stronger while maintaining stability. The coexistence of positive and negative effects of relevant elements in the land resource subsystem, especially influenced by the variation of salinization index, has curbed the trend of improving ecological security. The EFI values of both biological resource subsystem and environmental subsystem are significantly decreasing, and the ecological security level is significantly enhanced. On the basis of revealing the evolution process and laws of ecological security, this paper proposes ecological security risk prevention and control, pattern optimization, process regulation, ecological restoration and management models, providing support for promoting green, low-carbon, and high-quality development of river basins.
The Aksu River Basin landscape engineering, as a flagship project of the UN for ecological restoration, plays an important role in addressing climate change, protecting biodiversity, enhancing the value of ecological services and the quality of ecological products, and maintaining ecological security in the basin. Based on the law of regional differentiation, the principle of ecosystem coupling, and the mechanism of environmental effects, and based on the analysis of the coupling relationship between the spatiotemporal evolution of ecological vulnerability and ecological security, a ecological fragility evaluation index system for the Aksu River Basin"s mountains, rivers, forests, fields, lakes, grasslands, and sands system is constructed, consisting of 4 subsystems and 20 indicators. And apply the EFI method to analyze the contribution of various indicators and subsystems to EFI and their ecological effects.The larger the positive indicator value of EFI, the lower the ecological security.The larger the absolute value of the negative EFI indicator, the higher the ecological security. Obtained multi-source attribute data for various indicators in 1998 and 2023 through practical monitoring, experimental analysis, model simulation, remote sensing mapping, logical reasoning and empirical analysis, the weights and thresholds of each indicator are determined,which provide a data basis for ecological security evaluation through normalization and other methods. The results indicate that: (1) Ecological security is a complex and systematic issue. The interaction between various subsystems and their elements shows a certain degree of comprehensiveness, lag, complexity, and uncertainty in the ecological security of the basin. (2) The ecological security of the Aksu River Basin has been in a stable and improving state in recent 25 years. The EFI values are 0.08 and 0.06 in 1998 and 2023 respectively. According to the evaluation criteria, ecological fragility is further weakening and the degree of ecological security is continuously increasing. (3) The ecological security level of each subsystem and its contribution to the ecological security effect of the basin are different from each other. The indicators of the water resources subsystem are basically stable and gradually improving, and the fragility of the subsystem is becoming weaker while maintaining stability. The ecological security characteristics show an evolution pattern of becoming stronger while maintaining stability. The coexistence of positive and negative effects of relevant elements in the land resource subsystem, especially influenced by the variation of salinization index, has curbed the trend of improving ecological security. The EFI values of both biological resource subsystem and environmental subsystem are significantly decreasing, and the ecological security level is significantly enhanced. On the basis of revealing the evolution process and laws of ecological security, this paper proposes ecological security risk prevention and control, pattern optimization, process regulation, ecological restoration and management models, providing support for promoting green, low-carbon, and high-quality development of river basins.
Abstract:
In emotion recognition research, EEG, as a direct response to brain activity, can objectively reflect a person's emotional state. However, the non-smoothness and other characteristics of EEG signals make it more difficult to collect a large number of labelled EEG samples, thus limiting the effectiveness and generalisation performance of EEG emotion recognition methods to a certain extent. To address the above problems, a semi-supervised low-rank representation of the EEG emotion recognition method is proposed. Firstly, a regression form objective function is designed using the estimated labels of a small number of labelled EEG samples as a way to effectively estimate the labels of unlabelled samples. Second, a -drag-and-drop technique is used to ensure label-to-label separability, and in addition, low-rank constraints are imposed on the slack labels to improve their intra-class tightness and similarity. Thirdly, a class neighbourhood graph is incorporated into the proposed method as a way to capture the local neighbourhood information of all EEG sample data. The proposed method is examined on the SEED-IV and SEED-V datasets, and the results show that the proposed method has better performance on the EEG emotion recognition problems.
In emotion recognition research, EEG, as a direct response to brain activity, can objectively reflect a person's emotional state. However, the non-smoothness and other characteristics of EEG signals make it more difficult to collect a large number of labelled EEG samples, thus limiting the effectiveness and generalisation performance of EEG emotion recognition methods to a certain extent. To address the above problems, a semi-supervised low-rank representation of the EEG emotion recognition method is proposed. Firstly, a regression form objective function is designed using the estimated labels of a small number of labelled EEG samples as a way to effectively estimate the labels of unlabelled samples. Second, a -drag-and-drop technique is used to ensure label-to-label separability, and in addition, low-rank constraints are imposed on the slack labels to improve their intra-class tightness and similarity. Thirdly, a class neighbourhood graph is incorporated into the proposed method as a way to capture the local neighbourhood information of all EEG sample data. The proposed method is examined on the SEED-IV and SEED-V datasets, and the results show that the proposed method has better performance on the EEG emotion recognition problems.
Abstract:
Traditional numerical weather prediction (NWP) models have inherent biases in quantitative precipitation forecasting tasks due to limited spatial resolution, incomplete physical parameterization schemes, and weak generalization, Deep learning neural networks, with their strong non-linear fitting capabilities, autonomous learning of task-relevant features, and high generalization, hold the potential to address these issues and improve the current state of NWP forecasting. To address the aforementioned issues, this paper proposes a research on short-term quantitative precipitation forecasting technology based on multi factor 3D feature extraction. Based on high-resolution ECMWF HRES (EC Hs) model forecasting data provided by the European Centre for Medium Range Weather Forecasts (ECMWF), a 3D-QPF (3D-Quantitative Precision Forecast) semantic segmentation model is constructed. Through a coupled framework of classification and regression, the 3D spatial features of various precipitation related element data are captured, And obtain a non-linear relationship with the actual precipitation data, and finally add the PR (Pre and Rec) loss function to the loss function to further improve the predictive performance of the model on skewed data.The experimental results show that the accuracy score of 3D-QPF daily cumulative precipitation forecast not only increases steadily on the sunny and rainy scale (0.1mm/24h), but also improves significantly on the rainstorm scale (50mm/24h): the rainstorm scale is 15.8% higher than the EC-Hres 'TS(Threat Score) score, and the root mean square error (RMSE，Root Mean Square Error) optimization reaches 18.71%. After long-term testing, the 3D- QPF model has achieved effective prediction correction compared to classic network models such as ECHres, China Meteorological Administration Global Model (CMA-GFS) forecasting, 2D Unet, and 3D-Unet. In addition, as the forecast time is extended to 3 days, the optimization effect of the model can still remain relatively stable.
Traditional numerical weather prediction (NWP) models have inherent biases in quantitative precipitation forecasting tasks due to limited spatial resolution, incomplete physical parameterization schemes, and weak generalization, Deep learning neural networks, with their strong non-linear fitting capabilities, autonomous learning of task-relevant features, and high generalization, hold the potential to address these issues and improve the current state of NWP forecasting. To address the aforementioned issues, this paper proposes a research on short-term quantitative precipitation forecasting technology based on multi factor 3D feature extraction. Based on high-resolution ECMWF HRES (EC Hs) model forecasting data provided by the European Centre for Medium Range Weather Forecasts (ECMWF), a 3D-QPF (3D-Quantitative Precision Forecast) semantic segmentation model is constructed. Through a coupled framework of classification and regression, the 3D spatial features of various precipitation related element data are captured, And obtain a non-linear relationship with the actual precipitation data, and finally add the PR (Pre and Rec) loss function to the loss function to further improve the predictive performance of the model on skewed data.The experimental results show that the accuracy score of 3D-QPF daily cumulative precipitation forecast not only increases steadily on the sunny and rainy scale (0.1mm/24h), but also improves significantly on the rainstorm scale (50mm/24h): the rainstorm scale is 15.8% higher than the EC-Hres 'TS(Threat Score) score, and the root mean square error (RMSE，Root Mean Square Error) optimization reaches 18.71%. After long-term testing, the 3D- QPF model has achieved effective prediction correction compared to classic network models such as ECHres, China Meteorological Administration Global Model (CMA-GFS) forecasting, 2D Unet, and 3D-Unet. In addition, as the forecast time is extended to 3 days, the optimization effect of the model can still remain relatively stable.
Abstract:
In this study, a four-season field campaign was conducted from December 2017 to November 2018 for the study of aerosol optical properties in the northern suburbs of Nanjing. The aerosol absorption and scattering coefficients during the observation period were from high to low in winter, spring, fall, and summer, respectively, with bimodal daily variations, and the early peak time of absorption coefficients was early in summer and late in winter, which was due to the changes in the boundary layer heights and the timing of human production activities in different seasons. The average AAE was higher in summer and fall, indicating that brown carbon contributed more significantly to aerosol uptake, while black carbon uptake dominated in spring and winter.SAE was higher in spring and winter than in summer and fall, and both aerosols were dominated by fine modes, with smaller aerosol particle sizes in spring and winter. Aerosol optical properties are influenced by meteorological conditions and chemical components. The transport of pollutants under the influence of the northwest monsoon leads to high values of aerosol extinction coefficients during the fall and winter seasons when the wind is from the northwest. Local pollutants in the direction of east and southeast winds spread to the observation site with wind in spring and summer seasons, resulting in high values of aerosol extinction coefficients. The high aerosol extinction coefficients at both low and high wind speeds during the fall and winter seasons may be attributed to the accumulation of pollutants due to poor diffusion conditions at low wind speeds and the long-range transport of pollutants at high wind speeds. Aerosol extinction coefficients at the observation sites increase rapidly when air masses from inland are transported to the sites, and decrease when air masses from the sea are transported to the sites. When the relative humidity is less than 90%, the aerosol extinction coefficient is inversely proportional to the visibility, while when the relative humidity is more than 90%, the visibility is generally less than 10km, and the relationship with the extinction coefficient is not significant. The water-soluble ions in the PM2.5 components in the spring and winter seasons are mainly secondary source ions, and the correlation between aerosol extinction coefficients and nitrate ions is higher than that of sulfate ions, and pollutants affecting the aerosol extinction coefficients mainly come from transportation sources. In contrast, pollutants affecting aerosol extinction coefficients in summer and fall were mainly from stationary sources such as coal combustion. The contribution of brown carbon to the absorption coefficients was greater in summer than in fall, probably due to the increase in black carbon emissions in fall, which increased its light absorption contribution, and the relative decrease in the light absorption contribution of brown carbon.
In this study, a four-season field campaign was conducted from December 2017 to November 2018 for the study of aerosol optical properties in the northern suburbs of Nanjing. The aerosol absorption and scattering coefficients during the observation period were from high to low in winter, spring, fall, and summer, respectively, with bimodal daily variations, and the early peak time of absorption coefficients was early in summer and late in winter, which was due to the changes in the boundary layer heights and the timing of human production activities in different seasons. The average AAE was higher in summer and fall, indicating that brown carbon contributed more significantly to aerosol uptake, while black carbon uptake dominated in spring and winter.SAE was higher in spring and winter than in summer and fall, and both aerosols were dominated by fine modes, with smaller aerosol particle sizes in spring and winter. Aerosol optical properties are influenced by meteorological conditions and chemical components. The transport of pollutants under the influence of the northwest monsoon leads to high values of aerosol extinction coefficients during the fall and winter seasons when the wind is from the northwest. Local pollutants in the direction of east and southeast winds spread to the observation site with wind in spring and summer seasons, resulting in high values of aerosol extinction coefficients. The high aerosol extinction coefficients at both low and high wind speeds during the fall and winter seasons may be attributed to the accumulation of pollutants due to poor diffusion conditions at low wind speeds and the long-range transport of pollutants at high wind speeds. Aerosol extinction coefficients at the observation sites increase rapidly when air masses from inland are transported to the sites, and decrease when air masses from the sea are transported to the sites. When the relative humidity is less than 90%, the aerosol extinction coefficient is inversely proportional to the visibility, while when the relative humidity is more than 90%, the visibility is generally less than 10km, and the relationship with the extinction coefficient is not significant. The water-soluble ions in the PM2.5 components in the spring and winter seasons are mainly secondary source ions, and the correlation between aerosol extinction coefficients and nitrate ions is higher than that of sulfate ions, and pollutants affecting the aerosol extinction coefficients mainly come from transportation sources. In contrast, pollutants affecting aerosol extinction coefficients in summer and fall were mainly from stationary sources such as coal combustion. The contribution of brown carbon to the absorption coefficients was greater in summer than in fall, probably due to the increase in black carbon emissions in fall, which increased its light absorption contribution, and the relative decrease in the light absorption contribution of brown carbon.
Abstract:
In view of the fact that the existing person re-identification methods are difficult to avoid inaccurate feature extraction caused by environmental noise and are easily mistaken for person features, a person multi-feature fusion branch network based on dynamic convolution and attention mechanism is proposed. Firstly, due to the uncertain factors such as illumination change, human posture, object occlusion, etc., dynamic convolution is proposed to replace static convolution in ResNet50 to obtain a more robust Dy-ResNet50 model. Secondly, considering that there are great differences in the perspective of taking pictures of people and people are blocked by objects, it is proposed to embed the self-attention mechanism and the cross-attention mechanism into the backbone network. Finally, the cross entropy loss function and the difficult sample ternary loss function are used as the loss function together.Experiments are carried out on DukeMTMC-ReID、 Market-1501 and MSMT17 public datasets and compared with mainstream network models. The results show that the Rank-1 and mAP of the proposed model is 0.9 and 1.6 percentage points higher than that of the current mainstream model on the data set DukeMTMC-ReID.
In view of the fact that the existing person re-identification methods are difficult to avoid inaccurate feature extraction caused by environmental noise and are easily mistaken for person features, a person multi-feature fusion branch network based on dynamic convolution and attention mechanism is proposed. Firstly, due to the uncertain factors such as illumination change, human posture, object occlusion, etc., dynamic convolution is proposed to replace static convolution in ResNet50 to obtain a more robust Dy-ResNet50 model. Secondly, considering that there are great differences in the perspective of taking pictures of people and people are blocked by objects, it is proposed to embed the self-attention mechanism and the cross-attention mechanism into the backbone network. Finally, the cross entropy loss function and the difficult sample ternary loss function are used as the loss function together.Experiments are carried out on DukeMTMC-ReID、 Market-1501 and MSMT17 public datasets and compared with mainstream network models. The results show that the Rank-1 and mAP of the proposed model is 0.9 and 1.6 percentage points higher than that of the current mainstream model on the data set DukeMTMC-ReID.
Abstract:
As a crucial component of weather systems, 3D cloud simulation plays a significant role in various fields such as military and aviation. However, the current mainstream Bounding Volume Hierarchy (BVH) algorithm exhibits inefficient rendering performance when dealing with non-uniform and large-volume clouds. To address this issue, a cloud rendering method based on optimized BVH algorithm is proposed. The data points from the WRF grid are used as cloud primitives, and a Z-order Hilbert curve is employed for spatial sorting. The BVH algorithm based on the Surface Area Heuristic (SAH) is optimized by locally optimizing the cloud primitive density, aiming to enhance computational efficiency. To tackle the data access overhead of overlapping BVH nodes, a novel storage structure called Overlapping Node Sets (ONS) is introduced, which reduces the time complexity. The optimized BVH algorithm reduces unnecessary intersection tests between rays and triangle surfaces, and resolves issues related to invalid boundary volume overlaps. Simulation experiments demonstrate that the proposed method achieves a 15.6% improvement in SAH cost compared to similar state-of-the-art algorithms, a 10% improvement in EPO, and a reduction of over 100% in construction time. The computational efficiency of the optimized BVH algorithm outperforms similar algorithms in any WRF cloud scenario, indicating its capability for rapid rendering of WRF cloud products.
As a crucial component of weather systems, 3D cloud simulation plays a significant role in various fields such as military and aviation. However, the current mainstream Bounding Volume Hierarchy (BVH) algorithm exhibits inefficient rendering performance when dealing with non-uniform and large-volume clouds. To address this issue, a cloud rendering method based on optimized BVH algorithm is proposed. The data points from the WRF grid are used as cloud primitives, and a Z-order Hilbert curve is employed for spatial sorting. The BVH algorithm based on the Surface Area Heuristic (SAH) is optimized by locally optimizing the cloud primitive density, aiming to enhance computational efficiency. To tackle the data access overhead of overlapping BVH nodes, a novel storage structure called Overlapping Node Sets (ONS) is introduced, which reduces the time complexity. The optimized BVH algorithm reduces unnecessary intersection tests between rays and triangle surfaces, and resolves issues related to invalid boundary volume overlaps. Simulation experiments demonstrate that the proposed method achieves a 15.6% improvement in SAH cost compared to similar state-of-the-art algorithms, a 10% improvement in EPO, and a reduction of over 100% in construction time. The computational efficiency of the optimized BVH algorithm outperforms similar algorithms in any WRF cloud scenario, indicating its capability for rapid rendering of WRF cloud products.
Abstract:
Texture extraction has always been a crucial task in the field of computer vision, and the quality of texture extraction often has a critical impact on the accuracy of texture classification. Traditional single texture extraction methods struggle to accurately describe the characteristics of various textures. Therefore, this paper proposes a texture extraction algorithm based on an improved Position Local Binary Patterns (IPLBP) and Gabor filters. The improved algorithm enhances the texture description capability by extracting texture position information based on the Local Binary Patterns (LBP). In this paper, the improved algorithm is used to extract local texture information, while Gabor filters are employed to extract global texture information. The two types of feature information are then fused and classified using Support Vector Machines(SVM).Experimental results demonstrate that the proposed algorithm exhibits excellent performance on texture material classification tasks. Compared to traditional LBP algorithms, this algorithm can more accurately capture the differences between different texture features.
Texture extraction has always been a crucial task in the field of computer vision, and the quality of texture extraction often has a critical impact on the accuracy of texture classification. Traditional single texture extraction methods struggle to accurately describe the characteristics of various textures. Therefore, this paper proposes a texture extraction algorithm based on an improved Position Local Binary Patterns (IPLBP) and Gabor filters. The improved algorithm enhances the texture description capability by extracting texture position information based on the Local Binary Patterns (LBP). In this paper, the improved algorithm is used to extract local texture information, while Gabor filters are employed to extract global texture information. The two types of feature information are then fused and classified using Support Vector Machines(SVM).Experimental results demonstrate that the proposed algorithm exhibits excellent performance on texture material classification tasks. Compared to traditional LBP algorithms, this algorithm can more accurately capture the differences between different texture features.
Abstract:
In recent years, traffic flow prediction methods based on deep learning have been a hot research topic in the field of traffic flow prediction. Unlike traditional convolutional neural networks, graph convolutional networks, which are suitable for processing non-Euclidean data, have shown powerful capabilities in spatial feature modeling. Meanwhile, topological maps, distance maps and flow similarity maps that reflect the spatial characteristics of the road network are typical non-European data. Therefore, traffic flow prediction based on graph convolutional networks and its variants has become a research hotspot in the field of traffic flow prediction, and many attractive research results have been achieved. This article classifies and summarizes traffic flow prediction models based on graph convolutional networks in recent years. First, starting from the basic definition of graph convolutional network, the basic principles of graph convolution are described in detail by combining the definitions of spatial convolution and spectral convolution. Secondly, according to the network structure characteristics of the prediction model, the traffic flow prediction models based on graph convolutional network are divided into two major categories: "combined type" and "improved type". The most representative model structures are analyzed and discussed in detail. In addition, a review of typical data sets commonly used in the field of traffic flow prediction for model performance comparison is provided, and a simulation test is conducted using one of these real datasets to demonstrate the prediction performance of four traffic flow prediction models based on graph convolutional networks. Finally, based on the current research status and development trends, an open discussion and outlook on the future research hotspots and challenges in the field of traffic flow prediction methods based on graph convolutional networks are conducted.
In recent years, traffic flow prediction methods based on deep learning have been a hot research topic in the field of traffic flow prediction. Unlike traditional convolutional neural networks, graph convolutional networks, which are suitable for processing non-Euclidean data, have shown powerful capabilities in spatial feature modeling. Meanwhile, topological maps, distance maps and flow similarity maps that reflect the spatial characteristics of the road network are typical non-European data. Therefore, traffic flow prediction based on graph convolutional networks and its variants has become a research hotspot in the field of traffic flow prediction, and many attractive research results have been achieved. This article classifies and summarizes traffic flow prediction models based on graph convolutional networks in recent years. First, starting from the basic definition of graph convolutional network, the basic principles of graph convolution are described in detail by combining the definitions of spatial convolution and spectral convolution. Secondly, according to the network structure characteristics of the prediction model, the traffic flow prediction models based on graph convolutional network are divided into two major categories: "combined type" and "improved type". The most representative model structures are analyzed and discussed in detail. In addition, a review of typical data sets commonly used in the field of traffic flow prediction for model performance comparison is provided, and a simulation test is conducted using one of these real datasets to demonstrate the prediction performance of four traffic flow prediction models based on graph convolutional networks. Finally, based on the current research status and development trends, an open discussion and outlook on the future research hotspots and challenges in the field of traffic flow prediction methods based on graph convolutional networks are conducted.
Abstract:
After decades of development, laser and visual SLAM technology has been relatively mature and widely used in military and civil fields. The technology allows robots to move in indoor and outdoor scenarios where GPS signals are scarce. However, SLAM technology relying only on a single sensor has its own limitations. For example,laser SLAM is not suitable for scenes with a large number of dynamic objects around it, and visual SLAM has poor robustness in low-texture environments. However, the fusion of the two technologies has great potential to learn from each other. Therefore, this paper predicts that SLAM technology combining laser and vision or even more sensors will be the mainstream direction in the future. This paper reviews the development of SLAM technology, analyzes the hardware information of LIDAR and camera, and gives some classical open-source algorithms and datasets. According to the algorithm used in the fusion sensor, the multi-sensor fusion scheme is introduced in detail from the perspective of traditional uncertainty based, feature-based and novel deep learning based. The excellent performance of multi-sensor fusion scheme in complex scenes is summarized, and the future development of the multi-sensor fusion scheme is prospected.
After decades of development, laser and visual SLAM technology has been relatively mature and widely used in military and civil fields. The technology allows robots to move in indoor and outdoor scenarios where GPS signals are scarce. However, SLAM technology relying only on a single sensor has its own limitations. For example,laser SLAM is not suitable for scenes with a large number of dynamic objects around it, and visual SLAM has poor robustness in low-texture environments. However, the fusion of the two technologies has great potential to learn from each other. Therefore, this paper predicts that SLAM technology combining laser and vision or even more sensors will be the mainstream direction in the future. This paper reviews the development of SLAM technology, analyzes the hardware information of LIDAR and camera, and gives some classical open-source algorithms and datasets. According to the algorithm used in the fusion sensor, the multi-sensor fusion scheme is introduced in detail from the perspective of traditional uncertainty based, feature-based and novel deep learning based. The excellent performance of multi-sensor fusion scheme in complex scenes is summarized, and the future development of the multi-sensor fusion scheme is prospected.
Abstract:
Aiming at the predict problems of non-static, highly complex and random fluctuations of stock prices, a combination model based on variational mode decomposition(VMD)-circle sparrow search algorithm(CSSA)-long short-term memory neural network(LSTM) is established. Firstly, the original stock closing data is decomposed into several intrinsic mode function (IMF) components by VMD, and then the CSSA is used to optimize the parameters of hidden layer neurons, iteration number and learning rate of LSTM, and the optimal parameters are fitted into the LSTM, where each IMF component is modeled and predicted, and the prediction results of IMF component are superimposed to obtain the final result. Experiments show that the the different kinds of errors of the proposed model on multiple stock datasets are minimized, and the proposed model has better fitting in stock price and higher accuracy in complex stock price prediction.
Aiming at the predict problems of non-static, highly complex and random fluctuations of stock prices, a combination model based on variational mode decomposition(VMD)-circle sparrow search algorithm(CSSA)-long short-term memory neural network(LSTM) is established. Firstly, the original stock closing data is decomposed into several intrinsic mode function (IMF) components by VMD, and then the CSSA is used to optimize the parameters of hidden layer neurons, iteration number and learning rate of LSTM, and the optimal parameters are fitted into the LSTM, where each IMF component is modeled and predicted, and the prediction results of IMF component are superimposed to obtain the final result. Experiments show that the the different kinds of errors of the proposed model on multiple stock datasets are minimized, and the proposed model has better fitting in stock price and higher accuracy in complex stock price prediction.
Abstract:
An active noise control (ANC) method is introduced which replaces filtered-x least mean square (FxLMS) algorithm with Dual-Decoder Convolutional Recurrent Network (DCRN). Due to the importance of phase information in ANC, the input feature of the DCRN is the complex spectrogram of the noise signal (including the real and imaginary spectrograms). In the network structure, a coding module is used to extract features from the noise complex spectrograms, and a dual-decoder module is used to estimate the real and imaginary spectrograms of the network output respectively. Parameter sharing mechanism and group strategy are adopted to reduce the number of training parameters and improve the learning ability and generalization ability. Especially for wind noise, a new loss function is selected and the training data is regularized to improve the performance of DCRN. The experiments in both simulation and ANC headphone environments show that the DCRN method exhibits good noise reduction performance and robustness for general noise and wind noise.
An active noise control (ANC) method is introduced which replaces filtered-x least mean square (FxLMS) algorithm with Dual-Decoder Convolutional Recurrent Network (DCRN). Due to the importance of phase information in ANC, the input feature of the DCRN is the complex spectrogram of the noise signal (including the real and imaginary spectrograms). In the network structure, a coding module is used to extract features from the noise complex spectrograms, and a dual-decoder module is used to estimate the real and imaginary spectrograms of the network output respectively. Parameter sharing mechanism and group strategy are adopted to reduce the number of training parameters and improve the learning ability and generalization ability. Especially for wind noise, a new loss function is selected and the training data is regularized to improve the performance of DCRN. The experiments in both simulation and ANC headphone environments show that the DCRN method exhibits good noise reduction performance and robustness for general noise and wind noise.
Abstract:
Integrated energy systems enable the supply of multiple forms of energy, but at the same time emit a large amount of carbon dioxide, which also affects the surrounding environment. An optimal scheduling strategy based on double delay depth deterministic strategy gradient algorithm is proposed for low carbon economy scheduling in integrated energy systems. Firstly, taking the minimum operation cost as the objective function, a comprehensive energy system model with multiple complementary energies of electricity, heat and cold is established considering carbon capture technology and power to gas technology. Secondly, introduce a carbon trading mechanism to improve the enthusiasm of energy conservation and emission reduction in optimal scheduling. Then, according to the reinforcement learning framework, the state space, action space and reward function of the optimization model are designed, and the agents in the TD3 algorithm are used to interact with the environment to explore strategies and learn the operation strategy of the integrated energy system. Finally, we use the historical data to train the agent of TD3 algorithm, and compare the linear programming algorithm and particle swarm optimization algorithm in different scenarios. The results show that the proposed method can reduce the carbon emission and operating cost of the integrated energy system, and can realize the low-carbon economic dispatch of the integrated energy system.
Integrated energy systems enable the supply of multiple forms of energy, but at the same time emit a large amount of carbon dioxide, which also affects the surrounding environment. An optimal scheduling strategy based on double delay depth deterministic strategy gradient algorithm is proposed for low carbon economy scheduling in integrated energy systems. Firstly, taking the minimum operation cost as the objective function, a comprehensive energy system model with multiple complementary energies of electricity, heat and cold is established considering carbon capture technology and power to gas technology. Secondly, introduce a carbon trading mechanism to improve the enthusiasm of energy conservation and emission reduction in optimal scheduling. Then, according to the reinforcement learning framework, the state space, action space and reward function of the optimization model are designed, and the agents in the TD3 algorithm are used to interact with the environment to explore strategies and learn the operation strategy of the integrated energy system. Finally, we use the historical data to train the agent of TD3 algorithm, and compare the linear programming algorithm and particle swarm optimization algorithm in different scenarios. The results show that the proposed method can reduce the carbon emission and operating cost of the integrated energy system, and can realize the low-carbon economic dispatch of the integrated energy system.
Abstract:
Considering the problems of internal and external disturbances in permanent magnet synchronous motor(PMSM) including modeling errors, sudden load variation, etc, which affect the stability of speed control, a load estimation-based composite active disturbance rejection controller(ADRC) for PMSM speed regulation is proposed. ADRC is adopted to replace the PI controller in the speed loop for improving the performance of control system and solving the contradiction between system rapidity and overshooting. A load torque observer is designed to solve the problem of slow response to sudden load variation from ESO by directly estimating and compensating the load torque in real time via speed and current signals. Based on Matlab/Simulink environment, a semi-physical experimental platform of composite ADRC for speed regulation is constructed, and compared and verified with traditional PI control and linear ADRC. Promising results illustrated that the proposed load estimation-based composite ADRC reduces the speed variation by more than 30% compared with conventional controller when load variation abruptly occur, and has superior ability of disturbance rejection and speed regulation performance.
Considering the problems of internal and external disturbances in permanent magnet synchronous motor(PMSM) including modeling errors, sudden load variation, etc, which affect the stability of speed control, a load estimation-based composite active disturbance rejection controller(ADRC) for PMSM speed regulation is proposed. ADRC is adopted to replace the PI controller in the speed loop for improving the performance of control system and solving the contradiction between system rapidity and overshooting. A load torque observer is designed to solve the problem of slow response to sudden load variation from ESO by directly estimating and compensating the load torque in real time via speed and current signals. Based on Matlab/Simulink environment, a semi-physical experimental platform of composite ADRC for speed regulation is constructed, and compared and verified with traditional PI control and linear ADRC. Promising results illustrated that the proposed load estimation-based composite ADRC reduces the speed variation by more than 30% compared with conventional controller when load variation abruptly occur, and has superior ability of disturbance rejection and speed regulation performance.
Abstract:
For second-order leader-follower multi-agent systems with unknown nonlinear functions, an event-triggered r sliding mode control design scheme is proposed. This paper consists of two parts. In the first part, a new sliding mode reaching law based on inverse hyperbolic sine function is selected to ensure the consistency of multi-agent systems in finite time. In the second part, the sliding mode function with gain scaling factor r is designed and the event triggering mechanism is introduced. Through Lyapunov stability analysis, it is proved that the proposed control method is effective, which can not only eliminate the chattering in the system but also reduce the sampling frequency of the control action. In addition, the minimum lower bound of the triggering time interval is proved, which excludes Zeno phenomenon. Finally, the effectiveness of the proposed algorithm is verified by Matlab/simulink simulation results.
For second-order leader-follower multi-agent systems with unknown nonlinear functions, an event-triggered r sliding mode control design scheme is proposed. This paper consists of two parts. In the first part, a new sliding mode reaching law based on inverse hyperbolic sine function is selected to ensure the consistency of multi-agent systems in finite time. In the second part, the sliding mode function with gain scaling factor r is designed and the event triggering mechanism is introduced. Through Lyapunov stability analysis, it is proved that the proposed control method is effective, which can not only eliminate the chattering in the system but also reduce the sampling frequency of the control action. In addition, the minimum lower bound of the triggering time interval is proved, which excludes Zeno phenomenon. Finally, the effectiveness of the proposed algorithm is verified by Matlab/simulink simulation results.
Abstract:
This paper mainly addresses the problem of current sensorless finite time control for buck-boost converter with unknown constant power load. The negative impedance of constant power loads cause low frequency oscillation, which adversely affect the stability of the system. First, to reconstruct unavailable inductor current and unknown power load, a reduced-order generalized parameter estimation based observer with finite time convergence is developed on the basis of dynamic regressor extension and mixing technique, which is able to reformulate state observation as parameter estimation. Secondly, based on a feedback linearization method, this nonlinear system is transformed into a linear system and a fast terminal sliding mode controller is designed to stabilize the system. Subsequently, a current sensorless finite time controller for this system is proposed by combining a fast terminal sliding mode controller with a generalized parameter estimation based observer. Based on the finite time stability result of the cascaded system, the finite time stability of closed-loop system is proved. Finally, the effectiveness of the proposed current sensorless finite time control is verified by simulation and experiment results.
This paper mainly addresses the problem of current sensorless finite time control for buck-boost converter with unknown constant power load. The negative impedance of constant power loads cause low frequency oscillation, which adversely affect the stability of the system. First, to reconstruct unavailable inductor current and unknown power load, a reduced-order generalized parameter estimation based observer with finite time convergence is developed on the basis of dynamic regressor extension and mixing technique, which is able to reformulate state observation as parameter estimation. Secondly, based on a feedback linearization method, this nonlinear system is transformed into a linear system and a fast terminal sliding mode controller is designed to stabilize the system. Subsequently, a current sensorless finite time controller for this system is proposed by combining a fast terminal sliding mode controller with a generalized parameter estimation based observer. Based on the finite time stability result of the cascaded system, the finite time stability of closed-loop system is proved. Finally, the effectiveness of the proposed current sensorless finite time control is verified by simulation and experiment results.
Abstract:
A backtracking scrambled key generation algorithm is proposed to address the problems of slow channel change, the lack of randomness and security in quasi-static environments. A scrambling code is generated using the sum of the in-consistent indexes between the current key and the previous key and scrambled with the current key, resulting in a repetition ratio between successive generated keys close to 0.5 and passing the NIST test. Meanwhile, to eavesdrop on the current key, it is necessary not only to eavesdrop on the current channel coefficients, but also to successfully eavesdrop on the previous key to obtain the scrambled code for this encryption, however, eavesdropping on the previous key is the same as this. As the number of "one-time pad" key generation increases, even with the higher relevant channel coefficients overheard, the key leakage rate remains close to 0.5. The upper bounds on the probability of successful eavesdropping and its variation with the number of key generation N for general and bad channel conditions are estimated using semantic security and information-theoretic inequalities, respectively, giving certain parameters, the upper bounds for these two cases are obtained as and .
A backtracking scrambled key generation algorithm is proposed to address the problems of slow channel change, the lack of randomness and security in quasi-static environments. A scrambling code is generated using the sum of the in-consistent indexes between the current key and the previous key and scrambled with the current key, resulting in a repetition ratio between successive generated keys close to 0.5 and passing the NIST test. Meanwhile, to eavesdrop on the current key, it is necessary not only to eavesdrop on the current channel coefficients, but also to successfully eavesdrop on the previous key to obtain the scrambled code for this encryption, however, eavesdropping on the previous key is the same as this. As the number of "one-time pad" key generation increases, even with the higher relevant channel coefficients overheard, the key leakage rate remains close to 0.5. The upper bounds on the probability of successful eavesdropping and its variation with the number of key generation N for general and bad channel conditions are estimated using semantic security and information-theoretic inequalities, respectively, giving certain parameters, the upper bounds for these two cases are obtained as and .
Abstract:
In order to improve the accuracy of traditional data-driven machine learning models in short-term wind power prediction, a short-term wind power prediction model based on ikPCA-FABAS-KELM is proposed. Firstly, an improvement is made to principal component analysis, proposing reversible kernel principal component analysis (ikPCA), which reduces the complexity of input data while ensuring data features, in order to improve the running speed of the model; Secondly, the introduction of individual attraction strategies for fireflies improves the Tenebrio algorithm and proposes the FABAS algorithm; Finally, the FABAS algorithm is used to optimize the regularization parameter C and kernel parameters γ of the kernel limit learning machine, reducing the impact of manual parameter setting on blind model training and improving model prediction accuracy. The simulation results show that the proposed prediction model effectively improves the prediction accuracy of traditional models.
In order to improve the accuracy of traditional data-driven machine learning models in short-term wind power prediction, a short-term wind power prediction model based on ikPCA-FABAS-KELM is proposed. Firstly, an improvement is made to principal component analysis, proposing reversible kernel principal component analysis (ikPCA), which reduces the complexity of input data while ensuring data features, in order to improve the running speed of the model; Secondly, the introduction of individual attraction strategies for fireflies improves the Tenebrio algorithm and proposes the FABAS algorithm; Finally, the FABAS algorithm is used to optimize the regularization parameter C and kernel parameters γ of the kernel limit learning machine, reducing the impact of manual parameter setting on blind model training and improving model prediction accuracy. The simulation results show that the proposed prediction model effectively improves the prediction accuracy of traditional models.
Abstract:
To address the problem of low segmentation accuracy of multifracture muscle lesion sites in MRI images of patients with lumbar disc herniation, this paper proposes a new model to improve the U2net network with the goal that the encoding and decoding subnetworks are interconnected by a series of nested jump paths, and this redesigned jump connection is mainly used to reduce the semantic missing of feature maps in the encoding and decoding subnetworks. For this purpose, the jump connections in the middle of RSU-7, RSU-6,RSU-5,RSU-4 in the U2Net model are redesigned, and the RSU-4F part remains unchanged. In addition, the channel attention module was added to extract high quality multifractal muscle features, which effectively helps information flow in the network by using an attention mechanism to increase expressiveness, focus on important features and suppress unnecessary features. To verify the effectiveness of the model, experiments are conducted on the multi-cleft muscle MRI image dataset, and it is found that MIOU, DICE, and HD are improved compared with U-Net, U2-Net, and U-Net++ network structures. The experimental results show that the algorithm proposed in this paper has a good effect on MRI image segmentation of multifidus muscle, which can assist doctors to make judgments on the condition.
To address the problem of low segmentation accuracy of multifracture muscle lesion sites in MRI images of patients with lumbar disc herniation, this paper proposes a new model to improve the U2net network with the goal that the encoding and decoding subnetworks are interconnected by a series of nested jump paths, and this redesigned jump connection is mainly used to reduce the semantic missing of feature maps in the encoding and decoding subnetworks. For this purpose, the jump connections in the middle of RSU-7, RSU-6,RSU-5,RSU-4 in the U2Net model are redesigned, and the RSU-4F part remains unchanged. In addition, the channel attention module was added to extract high quality multifractal muscle features, which effectively helps information flow in the network by using an attention mechanism to increase expressiveness, focus on important features and suppress unnecessary features. To verify the effectiveness of the model, experiments are conducted on the multi-cleft muscle MRI image dataset, and it is found that MIOU, DICE, and HD are improved compared with U-Net, U2-Net, and U-Net++ network structures. The experimental results show that the algorithm proposed in this paper has a good effect on MRI image segmentation of multifidus muscle, which can assist doctors to make judgments on the condition.
Abstract:
In response to the problem that agricultural futures are affected by many factors so that nonlinear and nonsmooth features are difficult to extract and lead to low prediction accuracy, this paper proposes a forecasting method based on CEEMDAN and TransformerEncoder-TCN, which is from the idea of "decomposition-ensemble". First, the complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the time series into multiscale Intrinsic Mode Function (IMF) and residuals, reducing the complexity of series modeling. Second, each subseries is predicted using Temporal Convolutional Network (TCN) incorporating multi-stage self-attention unit (TransformerEncoder) which optimizes the modeling weights of the significant features of the sequence. Finally, the prediction results of each subseries are linearly summed and integrated to obtain the final prediction results. The soybean futures revenue index in the agricultural product index of South China Futures Company is used as the research object. The model is retrained by using time-series cross-validation and parameter transfer. The ablation and comparison experimental results show that the model proposed in this paper has superiority in RMSE, MAE and DS, verifying its effectiveness in predicting agricultural product futures.
In response to the problem that agricultural futures are affected by many factors so that nonlinear and nonsmooth features are difficult to extract and lead to low prediction accuracy, this paper proposes a forecasting method based on CEEMDAN and TransformerEncoder-TCN, which is from the idea of "decomposition-ensemble". First, the complementary ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the time series into multiscale Intrinsic Mode Function (IMF) and residuals, reducing the complexity of series modeling. Second, each subseries is predicted using Temporal Convolutional Network (TCN) incorporating multi-stage self-attention unit (TransformerEncoder) which optimizes the modeling weights of the significant features of the sequence. Finally, the prediction results of each subseries are linearly summed and integrated to obtain the final prediction results. The soybean futures revenue index in the agricultural product index of South China Futures Company is used as the research object. The model is retrained by using time-series cross-validation and parameter transfer. The ablation and comparison experimental results show that the model proposed in this paper has superiority in RMSE, MAE and DS, verifying its effectiveness in predicting agricultural product futures.
Abstract:
Federated learning is an important method for addressing two critical challenges in machine learning: data sharing and privacy protection. However, federated learning itself faces challenges related to data heterogeneity and model heterogeneity. Existing research often focuses on addressing one of these issues, overlooking the correlation between them. To address this, this paper introduces a framework named PFKD. This framework utilizes knowledge distillation techniques to address model heterogeneity and personalized algorithms to tackle data heterogeneity, thereby achieving more personalized federated learning. Experimental analysis validates the effectiveness of the pro-posed framework. The experimental results demonstrate that the framework can overcome model performance bottlenecks, improving model accuracy by approximately 1 percentage point. Furthermore, with the appropriate hyperparameter adjustments, the framework's performance is further enhanced, resulting in an increase in model accuracy by approximately 2 percentage points.
Federated learning is an important method for addressing two critical challenges in machine learning: data sharing and privacy protection. However, federated learning itself faces challenges related to data heterogeneity and model heterogeneity. Existing research often focuses on addressing one of these issues, overlooking the correlation between them. To address this, this paper introduces a framework named PFKD. This framework utilizes knowledge distillation techniques to address model heterogeneity and personalized algorithms to tackle data heterogeneity, thereby achieving more personalized federated learning. Experimental analysis validates the effectiveness of the pro-posed framework. The experimental results demonstrate that the framework can overcome model performance bottlenecks, improving model accuracy by approximately 1 percentage point. Furthermore, with the appropriate hyperparameter adjustments, the framework's performance is further enhanced, resulting in an increase in model accuracy by approximately 2 percentage points.
Abstract:
Aiming at the problem that a large number of existing artificial intelligence algorithms are difficult to detect stably due to the complex attitude and variable scale of marine targets, a detection algorithm based on depth supervision and improved YOLOv8 is proposed. Firstly, a multi-scale convolution module is designed to extract the feature information of the target's multi-receptive fields and reduce the missed detection rate. Then, a deep supervision network is added to improve the utilization ratio of deep class information and shallow location information, and optimize the performance of target feature extraction of the backbone network. Finally, a channel attention mechanism is introduced in the detection head to filter the irrelevant information and enhance the recognition rate of the key features. The experimental results in the marine target dataset show that the mAP value of the improved algorithm reaches 93.69% and the recall rate reaches 85.16%, which is 7.38 and 8.52 percentage points higher than the original model respectively, and is better than the classical algorithm and the novel algorithm, The detection time is about 14ms, which can meet the requirements of real-time target detection at marine and provide effective technical references for shipping management and marine accident prevention.
Aiming at the problem that a large number of existing artificial intelligence algorithms are difficult to detect stably due to the complex attitude and variable scale of marine targets, a detection algorithm based on depth supervision and improved YOLOv8 is proposed. Firstly, a multi-scale convolution module is designed to extract the feature information of the target's multi-receptive fields and reduce the missed detection rate. Then, a deep supervision network is added to improve the utilization ratio of deep class information and shallow location information, and optimize the performance of target feature extraction of the backbone network. Finally, a channel attention mechanism is introduced in the detection head to filter the irrelevant information and enhance the recognition rate of the key features. The experimental results in the marine target dataset show that the mAP value of the improved algorithm reaches 93.69% and the recall rate reaches 85.16%, which is 7.38 and 8.52 percentage points higher than the original model respectively, and is better than the classical algorithm and the novel algorithm, The detection time is about 14ms, which can meet the requirements of real-time target detection at marine and provide effective technical references for shipping management and marine accident prevention.
Abstract:
The existing remote sensing image object detection algorithms have the problem of large parameter quantities and difficulty in deploying on mobile devices. Therefore, this paper proposes a lightweight remote sensing image object detection algorithm without anchor boxes. Firstly, the DWS-Sandglass lightweight module was designed to reduce the model volume, and then the activation function of the model was improved to ensure the detection accuracy. Then, a parameter free attention module SimAM is introduced to enable the network to focus on more important feature information. Finally, prune the redundant channels of the anchor free box algorithm to reduce the number of model parameters, and fine tune to improve accuracy. The experimental results on the HRSC2016 dataset show that compared with the current mainstream anchor free frame detection algorithms, this algorithm has faster detection speed and smaller model size, making it more suitable for deployment on mobile devices with comparable detection accuracy.
The existing remote sensing image object detection algorithms have the problem of large parameter quantities and difficulty in deploying on mobile devices. Therefore, this paper proposes a lightweight remote sensing image object detection algorithm without anchor boxes. Firstly, the DWS-Sandglass lightweight module was designed to reduce the model volume, and then the activation function of the model was improved to ensure the detection accuracy. Then, a parameter free attention module SimAM is introduced to enable the network to focus on more important feature information. Finally, prune the redundant channels of the anchor free box algorithm to reduce the number of model parameters, and fine tune to improve accuracy. The experimental results on the HRSC2016 dataset show that compared with the current mainstream anchor free frame detection algorithms, this algorithm has faster detection speed and smaller model size, making it more suitable for deployment on mobile devices with comparable detection accuracy.
Abstract:
Abstract Based on sensor network transmission, the event-triggered distributed filtering problem for a class of linear systems with additive and multiplicative noises is investigated in this paper, in which the considered process noise and measurement noise have one-step autocorrelation and two-step cross-correlation characteristics. Firstly, a recursive equation is used to describe the dynamic deviation of the system and the random variables of the Bernoulli distribution are introduced to describe the random packet loss phenomenon.Secondly, an event-triggered mechanism is introduced to reduce the frequency of information transmission while ensuring filtering performance, and a novel distributed filter is constructed based on consistency.? Then, a recursive equation for the upper bound of filtering error covariance is established using random analysis technology, and an expression for filtering gain is given by minimizing the variance constraint index. Finally, the effectiveness of the proposed optimized filtering method was verified through numerical simulation.
Abstract Based on sensor network transmission, the event-triggered distributed filtering problem for a class of linear systems with additive and multiplicative noises is investigated in this paper, in which the considered process noise and measurement noise have one-step autocorrelation and two-step cross-correlation characteristics. Firstly, a recursive equation is used to describe the dynamic deviation of the system and the random variables of the Bernoulli distribution are introduced to describe the random packet loss phenomenon.Secondly, an event-triggered mechanism is introduced to reduce the frequency of information transmission while ensuring filtering performance, and a novel distributed filter is constructed based on consistency.? Then, a recursive equation for the upper bound of filtering error covariance is established using random analysis technology, and an expression for filtering gain is given by minimizing the variance constraint index. Finally, the effectiveness of the proposed optimized filtering method was verified through numerical simulation.
Abstract:
Aiming at the distributed permutation flow shop scheduling problem with variable processing speed, a double population algorithm is proposed to optimize the maximum completion time and the total energy consumption of the machine. Firstly, an initialization method that mixes four strategies is used to generate a high-quality initial population. Secondly, specific evolution methods are designed according to the characteristics of the two pop-ulations, and the dynamic guide factor is introduced to adjust the evolution mode of the population. At the same time, an energy-saving strategy for speed regulation is proposed to further optimize energy consumption. Finally, a dynamic population strategy is proposed to balance the resources of the two populations. The effectiveness of each strategy is proved by simulation experiments, and compared with other algorithms, the results show that the proposed algorithm has obvious advantages.
Aiming at the distributed permutation flow shop scheduling problem with variable processing speed, a double population algorithm is proposed to optimize the maximum completion time and the total energy consumption of the machine. Firstly, an initialization method that mixes four strategies is used to generate a high-quality initial population. Secondly, specific evolution methods are designed according to the characteristics of the two pop-ulations, and the dynamic guide factor is introduced to adjust the evolution mode of the population. At the same time, an energy-saving strategy for speed regulation is proposed to further optimize energy consumption. Finally, a dynamic population strategy is proposed to balance the resources of the two populations. The effectiveness of each strategy is proved by simulation experiments, and compared with other algorithms, the results show that the proposed algorithm has obvious advantages.
Abstract:
ABSRTACT: Aiming at the problems of high time cost, labor cost and low efficiency in traditional remote sensing image processing, and aiming at improving the processing speed, accuracy and reducing the number of parameters in the classification of remote sensing hyperspectral images (HSI), an improved 2DCNN model En-De-2CP-2DCNN is proposed.Firstly, 1DCNN, 2DCNN and 3DCNN were used to carry out classification experiments on the Pavia University HSI dataset, and the advantages and disadvantages of each were compared and analyzed.Secondly, under the premise of maintaining fast processing speed and not increasing the number of model parameters, 2DCNN is selected as the basic model, referring to the Encode-Decode structure of SegNet, and integrating the idea of double convolutional pooling to improve the basic model and optimize the learning strategy.The results show that the accuracy F1-score of the En-De-2CP-2DCNN model is 99.36%, which reaches the same level as 3DCNN (99.21%), which is 2.11% higher than that before the improvement (97.25%). The processing speed (7s/epoch) and 1DCNN are in the same order of magnitude, faster than 3DCNN (100s/epoch); The amount of parameters (3.66 MB) is higher than 3DCNN (318 KB), but much lower than 1DCNN (19.3 MB).It basically realizes the goal of accurate, fast and lightweight remote sensing hyperspectral image processing, and provides technical support for the automatic processing of remote sensing images.In particular, the improvement in processing speed and parameter amount is conducive to further realizing the lightweight deployment of mobile terminals..
ABSRTACT: Aiming at the problems of high time cost, labor cost and low efficiency in traditional remote sensing image processing, and aiming at improving the processing speed, accuracy and reducing the number of parameters in the classification of remote sensing hyperspectral images (HSI), an improved 2DCNN model En-De-2CP-2DCNN is proposed.Firstly, 1DCNN, 2DCNN and 3DCNN were used to carry out classification experiments on the Pavia University HSI dataset, and the advantages and disadvantages of each were compared and analyzed.Secondly, under the premise of maintaining fast processing speed and not increasing the number of model parameters, 2DCNN is selected as the basic model, referring to the Encode-Decode structure of SegNet, and integrating the idea of double convolutional pooling to improve the basic model and optimize the learning strategy.The results show that the accuracy F1-score of the En-De-2CP-2DCNN model is 99.36%, which reaches the same level as 3DCNN (99.21%), which is 2.11% higher than that before the improvement (97.25%). The processing speed (7s/epoch) and 1DCNN are in the same order of magnitude, faster than 3DCNN (100s/epoch); The amount of parameters (3.66 MB) is higher than 3DCNN (318 KB), but much lower than 1DCNN (19.3 MB).It basically realizes the goal of accurate, fast and lightweight remote sensing hyperspectral image processing, and provides technical support for the automatic processing of remote sensing images.In particular, the improvement in processing speed and parameter amount is conducive to further realizing the lightweight deployment of mobile terminals..
Abstract:
Computational offloading is an essential technology in mobile edge computing, and for the shortage of computational offloading strategies in multi-user multi-MEC server scenarios, this paper proposes a hybrid artificial bee colony algorithm (artificial reverse sine-cosine, ARSC). the ARSC algorithm first initializes the population using a Opposition-Based Learning strategy to optimize the initial solution of the population; then in the employment bee stage by using the sine-cosine algorithm of global optimal bootstrap information to improve the global search capability of the algorithm, and finally, to balance the global search capability and local search capability of the algorithm, the step size factor of the algorithm is improved by combining convex and concave functions. The offloading strategies based on particle swarm algorithm and artificial bee colony algorithm are contrasted with the ARSC approach through simulated studies. The experimental findings demonstrate that the convergence, system latency, and system energy consumption have all improved with the ARSC method.
Computational offloading is an essential technology in mobile edge computing, and for the shortage of computational offloading strategies in multi-user multi-MEC server scenarios, this paper proposes a hybrid artificial bee colony algorithm (artificial reverse sine-cosine, ARSC). the ARSC algorithm first initializes the population using a Opposition-Based Learning strategy to optimize the initial solution of the population; then in the employment bee stage by using the sine-cosine algorithm of global optimal bootstrap information to improve the global search capability of the algorithm, and finally, to balance the global search capability and local search capability of the algorithm, the step size factor of the algorithm is improved by combining convex and concave functions. The offloading strategies based on particle swarm algorithm and artificial bee colony algorithm are contrasted with the ARSC approach through simulated studies. The experimental findings demonstrate that the convergence, system latency, and system energy consumption have all improved with the ARSC method.
Abstract:
Aiming at the problems of low efficiency of manual observation and low recognition accuracy of crop development stage in China, a wheat development stage recognition method based on I_CBAM-DenseNet model is proposed. The method takes a dense connected convolutional network (DenseNet) as the backbone extraction network and incorporates a convolutional attention module CBAM. firstly, the spatial attention module (SAM) and channel attention module (CAM) in CBAM are changed from the traditional serial connection to parallel connection, and the improved CBAM (I_CBAM) is inserted into the last dense network of DenseNet. An I_CBAM-DenseNet model is constructed, and then seven important development periods of wheat are selected for automatic identification. In order to maximize the extraction of the feature information of wheat, the ultra-green feature of ExG factor and the the maximum inter-class variance method are combined to segment the acquired wheat images. The changes in accuracy and loss values of the models I_CBAM-DenseNet, AlexNet, ResNet, DenseNet, CBAM-DenseNet and VGG are comparatively analysed. The results show that the model built by taking the convolutional neural network based on I_CBAM-DenseNet has an accuracy of 99.64%, which is higher than the comparison models.
Aiming at the problems of low efficiency of manual observation and low recognition accuracy of crop development stage in China, a wheat development stage recognition method based on I_CBAM-DenseNet model is proposed. The method takes a dense connected convolutional network (DenseNet) as the backbone extraction network and incorporates a convolutional attention module CBAM. firstly, the spatial attention module (SAM) and channel attention module (CAM) in CBAM are changed from the traditional serial connection to parallel connection, and the improved CBAM (I_CBAM) is inserted into the last dense network of DenseNet. An I_CBAM-DenseNet model is constructed, and then seven important development periods of wheat are selected for automatic identification. In order to maximize the extraction of the feature information of wheat, the ultra-green feature of ExG factor and the the maximum inter-class variance method are combined to segment the acquired wheat images. The changes in accuracy and loss values of the models I_CBAM-DenseNet, AlexNet, ResNet, DenseNet, CBAM-DenseNet and VGG are comparatively analysed. The results show that the model built by taking the convolutional neural network based on I_CBAM-DenseNet has an accuracy of 99.64%, which is higher than the comparison models.
Abstract:
Phytoplankton is a major participant in the material and energy cycles of lake ecosystems, and obtaining information on its community structure is of great significance in coping with and regulating lake ecosystems. In this study, we examined the community characteristics of eukaryotic phytoplankton in the western half of Chaohu Lake in winter and summer by high-throughput sequencing. A total of 7 phyla and 71 Genus of eukaryotic phytoplankton were detected during the survey, 7 phyla and 59 Genus in summer and 5 phyla and 27 genera in winter, dominated by Chlorophyta and Bacillariophyta, and the dominant genus varied greatly in winter compared with those in summer. The mean values of Shannon-Wiener index in summer and winter were 1.83 and 1.88 respectively, and the mean values of Pielou index in summer and winter were 0.75 and 0.83 respectively. The mean values of both diversity indices were smaller in summer than in winter. The results of water quality analysis indicated that TN and TP were relatively high in the western half of Chaohu Lake during the study period, and the physicochemical of the water body were significantly different between summer and winter (p≤0.05). Redundancy analysis showed that PO4--P concentration, TN concentration, TP concentration and NH4+-N concentration had the greatest explanatory rate for the eukaryotic phytoplankton community, the effect of PO4--P concentration reaching a significant level (p≤0.05). Mantel correlation analysis showed that there was a good correlation between eukaryotic phytoplankton abundance with WT, DO, pH, NH4+-N, TN and Chl.a. Variance Partitioning Analysis showed that seasonal factors explained the eukaryotic phytoplankton community the most.
Phytoplankton is a major participant in the material and energy cycles of lake ecosystems, and obtaining information on its community structure is of great significance in coping with and regulating lake ecosystems. In this study, we examined the community characteristics of eukaryotic phytoplankton in the western half of Chaohu Lake in winter and summer by high-throughput sequencing. A total of 7 phyla and 71 Genus of eukaryotic phytoplankton were detected during the survey, 7 phyla and 59 Genus in summer and 5 phyla and 27 genera in winter, dominated by Chlorophyta and Bacillariophyta, and the dominant genus varied greatly in winter compared with those in summer. The mean values of Shannon-Wiener index in summer and winter were 1.83 and 1.88 respectively, and the mean values of Pielou index in summer and winter were 0.75 and 0.83 respectively. The mean values of both diversity indices were smaller in summer than in winter. The results of water quality analysis indicated that TN and TP were relatively high in the western half of Chaohu Lake during the study period, and the physicochemical of the water body were significantly different between summer and winter (p≤0.05). Redundancy analysis showed that PO4--P concentration, TN concentration, TP concentration and NH4+-N concentration had the greatest explanatory rate for the eukaryotic phytoplankton community, the effect of PO4--P concentration reaching a significant level (p≤0.05). Mantel correlation analysis showed that there was a good correlation between eukaryotic phytoplankton abundance with WT, DO, pH, NH4+-N, TN and Chl.a. Variance Partitioning Analysis showed that seasonal factors explained the eukaryotic phytoplankton community the most.
Abstract:
The electrical power output of photovoltaic array has multi-peak characteristics under partial shading conditions,and changes with the external environment.To achieve efficient electricity output,the Multi-Verse optimization (MVO) algorithm which has outstanding advantages in solving low dimensional and small-scale optimization problems is used to carry out maximum power point tracking (MPPT), and integrating multiple strategies to improve the defects,which increases the diversity of the universe population by initializing the universe population by Latin hypercube sampling sampling strategy and implementing Cauchy mutation on the universe randomly swapped according to roulette strategy.Meanwhile,the Levy Flying Quantum Particle swarm optimization (QPSO) algorithm is introduced, and the wormhole existence probability and travel distance rate are adaptively adjusted to enhance the global exploration and local development capabilities of the algorithm.The Matlab simulation results show that the algorithm reduces MPPT time by more than 45% and get higher MPPT accuracy compared with other algorithms,so it has better MPPT performance,and can improve the efficiency of photovoltaic power generation efficiency.
The electrical power output of photovoltaic array has multi-peak characteristics under partial shading conditions,and changes with the external environment.To achieve efficient electricity output,the Multi-Verse optimization (MVO) algorithm which has outstanding advantages in solving low dimensional and small-scale optimization problems is used to carry out maximum power point tracking (MPPT), and integrating multiple strategies to improve the defects,which increases the diversity of the universe population by initializing the universe population by Latin hypercube sampling sampling strategy and implementing Cauchy mutation on the universe randomly swapped according to roulette strategy.Meanwhile,the Levy Flying Quantum Particle swarm optimization (QPSO) algorithm is introduced, and the wormhole existence probability and travel distance rate are adaptively adjusted to enhance the global exploration and local development capabilities of the algorithm.The Matlab simulation results show that the algorithm reduces MPPT time by more than 45% and get higher MPPT accuracy compared with other algorithms,so it has better MPPT performance,and can improve the efficiency of photovoltaic power generation efficiency.
Abstract:
To investigate the response of N2O emission to the gradually elevated CO2 concentration and the reduction of nitrogen addition in wheat fields, we conducted field experiments using Yangmai 22 as the experimental material. The experiments were carried out on a field equipped with an open-top gas chamber, forming an automatic CO2 concentration control platform. This setup allowed us to carefully monitor and control the CO2 levels throughout the experiment. A gradually elevated CO2 concentration (EC, an increase of 40 μmol.mol-1 year by year from 2016-2017 wheat growing season until 120 μmol.mol-1 in 2018-2019 growing season) was set up, compared to the ambient atmospheric CO2 concentration (AC). The amount of nitrogen fertilizer were set as the conventional nitrogen addition (N1, 25 g.m-2) and the reduction treatment (N2, 15 g.m-2). The sampling and flux determination of N2O were carried out by static camera obscura gas chromatography in wheat field. The results indicated that throughout the winter wheat growing period, the trend of N2O emission under different CO2 concentration and nitrogen fertilizer treatment was consistent, and the overall trend showed a fluctuating decline. The impact of gradually elevated CO2 concentration on N2O emission was found to be not significant during the winter wheat growing season. The nitrogen reduction treatment has shown effective in reducing N2O emission in wheat fields, resulting in a significant reduction of cumulative N2O emission by 45.2% (p=0.004) under atmospheric CO2 concentration. The impact of nitrogen reduction treatment on N2O emission was evident in winter wheat from the booting to milking stage. Under the combined effect of gradually elevated CO2 concentration and reduction of nitrogen fertilizer addition, nitrogen addition is the main factor affecting N2O emission in wheat fields.
To investigate the response of N2O emission to the gradually elevated CO2 concentration and the reduction of nitrogen addition in wheat fields, we conducted field experiments using Yangmai 22 as the experimental material. The experiments were carried out on a field equipped with an open-top gas chamber, forming an automatic CO2 concentration control platform. This setup allowed us to carefully monitor and control the CO2 levels throughout the experiment. A gradually elevated CO2 concentration (EC, an increase of 40 μmol.mol-1 year by year from 2016-2017 wheat growing season until 120 μmol.mol-1 in 2018-2019 growing season) was set up, compared to the ambient atmospheric CO2 concentration (AC). The amount of nitrogen fertilizer were set as the conventional nitrogen addition (N1, 25 g.m-2) and the reduction treatment (N2, 15 g.m-2). The sampling and flux determination of N2O were carried out by static camera obscura gas chromatography in wheat field. The results indicated that throughout the winter wheat growing period, the trend of N2O emission under different CO2 concentration and nitrogen fertilizer treatment was consistent, and the overall trend showed a fluctuating decline. The impact of gradually elevated CO2 concentration on N2O emission was found to be not significant during the winter wheat growing season. The nitrogen reduction treatment has shown effective in reducing N2O emission in wheat fields, resulting in a significant reduction of cumulative N2O emission by 45.2% (p=0.004) under atmospheric CO2 concentration. The impact of nitrogen reduction treatment on N2O emission was evident in winter wheat from the booting to milking stage. Under the combined effect of gradually elevated CO2 concentration and reduction of nitrogen fertilizer addition, nitrogen addition is the main factor affecting N2O emission in wheat fields.
Abstract:
The development mode of transportation infrastructure construction has shifted from pursuing speed and scale to prioritizing quality and efficiency. As a result, there is an increased demand for fine-tuning and adapting the tunnel lighting environment. The design of the tunnel lighting system should prioritize the safety and comfort of drivers. Therefore, it is necessary to quantify the impact of tunnel lighting on driver safety and comfort during tunnel driving. Based on the theory of tunnel lighting and driver research, eight indicators that affect driver safety and comfort had been selected to establish a driver safety and comfort evaluation system. Indoor simulation research was conducted through a tunnel driving simulation environment and hardware facilities. A virtual simulation model of the tunnel was established using UC-win/Road software, and data corresponding to the eight indicators were obtained through driving simulators and physiological instruments. After processing the data, the entropy method was used to determine the weights of the indicators, and the safety and comfort of the tunnel under different lighting brightness and color temperature environments were evaluated. The target value of brightness and color temperature in the tunnel were obtained with the goal of achieving the optimal value of driver safety and comfort evaluation. The experimental results confirm the effectiveness of the driver safety and comfort evaluation system, propose an improved lighting scheme, and provide reference for tunnel operators.
The development mode of transportation infrastructure construction has shifted from pursuing speed and scale to prioritizing quality and efficiency. As a result, there is an increased demand for fine-tuning and adapting the tunnel lighting environment. The design of the tunnel lighting system should prioritize the safety and comfort of drivers. Therefore, it is necessary to quantify the impact of tunnel lighting on driver safety and comfort during tunnel driving. Based on the theory of tunnel lighting and driver research, eight indicators that affect driver safety and comfort had been selected to establish a driver safety and comfort evaluation system. Indoor simulation research was conducted through a tunnel driving simulation environment and hardware facilities. A virtual simulation model of the tunnel was established using UC-win/Road software, and data corresponding to the eight indicators were obtained through driving simulators and physiological instruments. After processing the data, the entropy method was used to determine the weights of the indicators, and the safety and comfort of the tunnel under different lighting brightness and color temperature environments were evaluated. The target value of brightness and color temperature in the tunnel were obtained with the goal of achieving the optimal value of driver safety and comfort evaluation. The experimental results confirm the effectiveness of the driver safety and comfort evaluation system, propose an improved lighting scheme, and provide reference for tunnel operators.
Abstract:
In the islanded microgrid, the mismatch in line impedance hinders the traditional droop control from achieving an equal distribution of reactive power among distributed generation (DG) sources. To achieve reactive power sharing among DG sources and enhance the flexibility and reliability of the controller, the paper first analyzes the reasons behind the failure of traditional droop control in achieving reactive power sharing and proposes an adaptive droop coefficient that can be dynamically adjusted to meet the conditions for reactive power sharing, thus resolving the issue of unequal reactive power distribution. Additionally, to enhance the flexibility and reliability of the reactive power sharing controller, a dynamic distributed observer is designed and its convergence is proven. The dynamic distributed observer allows DG sources to obtain the necessary information in a flexible and reliable manner in a distributed fashion. Finally, the proposed control strategy is validated through simulations in four different scenarios, demonstrating its superiority and effectiveness.
In the islanded microgrid, the mismatch in line impedance hinders the traditional droop control from achieving an equal distribution of reactive power among distributed generation (DG) sources. To achieve reactive power sharing among DG sources and enhance the flexibility and reliability of the controller, the paper first analyzes the reasons behind the failure of traditional droop control in achieving reactive power sharing and proposes an adaptive droop coefficient that can be dynamically adjusted to meet the conditions for reactive power sharing, thus resolving the issue of unequal reactive power distribution. Additionally, to enhance the flexibility and reliability of the reactive power sharing controller, a dynamic distributed observer is designed and its convergence is proven. The dynamic distributed observer allows DG sources to obtain the necessary information in a flexible and reliable manner in a distributed fashion. Finally, the proposed control strategy is validated through simulations in four different scenarios, demonstrating its superiority and effectiveness.
Abstract:
The lack of curbstones on the national and provincial highways makes it challenging to determine the edges of highways in point cloud scenarios, making it difficult to extract point clouds on national and provincial highways accurately. Therefore, a method for extracting point clouds on national and provincial highways is proposed by combining shape feature vector screening with local roughness. Firstly, a grid index is constructed for the original point cloud data. Based on the grid as the basic unit, a shape feature vector analysis method based on a covariance matrix is designed. Based on this method, the shape feature vectors of the point cloud within the unit are calculated, and an orthogonal basis is constructed to solve the feature values, filtering and obtaining the ground point cloud. Secondly, the KD tree index is used to reconstruct the ground point cloud, and the local roughness of the point cloud is calculated and analyzed. Based on a prior threshold of roughness, the Euclidean clustering algorithm is introduced to extract the point cloud of the road surface and its surrounding areas; Implement effective division between road points and non-road points. Testing point cloud data on 10 km mobile laser scanning of national and provincial main roads showed an average accuracy of 99.64% and a recall rate of 99.59%, respectively. Compared with the European clustering segmentation algorithm, it has improved the average accuracy and F1-score by 8.95% and 4.56%, respectively. Regarding computational efficiency, the algorithm has a processing speed of 32.4 km/h for the original point cloud, significantly improving efficiency, and has good road surface extraction accuracy and robustness.
The lack of curbstones on the national and provincial highways makes it challenging to determine the edges of highways in point cloud scenarios, making it difficult to extract point clouds on national and provincial highways accurately. Therefore, a method for extracting point clouds on national and provincial highways is proposed by combining shape feature vector screening with local roughness. Firstly, a grid index is constructed for the original point cloud data. Based on the grid as the basic unit, a shape feature vector analysis method based on a covariance matrix is designed. Based on this method, the shape feature vectors of the point cloud within the unit are calculated, and an orthogonal basis is constructed to solve the feature values, filtering and obtaining the ground point cloud. Secondly, the KD tree index is used to reconstruct the ground point cloud, and the local roughness of the point cloud is calculated and analyzed. Based on a prior threshold of roughness, the Euclidean clustering algorithm is introduced to extract the point cloud of the road surface and its surrounding areas; Implement effective division between road points and non-road points. Testing point cloud data on 10 km mobile laser scanning of national and provincial main roads showed an average accuracy of 99.64% and a recall rate of 99.59%, respectively. Compared with the European clustering segmentation algorithm, it has improved the average accuracy and F1-score by 8.95% and 4.56%, respectively. Regarding computational efficiency, the algorithm has a processing speed of 32.4 km/h for the original point cloud, significantly improving efficiency, and has good road surface extraction accuracy and robustness.
Abstract:
In order to solve the problem that the poor image quality leads to the low efficiency of subsequent machine vision tasks in rainy days, an image rain removal algorithm based on multi-feature interaction and dense residual is proposed. Firstly, a multi-feature interactive convolution module is proposed to extract the semantic features of rain lines in different spaces in order to improve the utilization of information. Secondly, a multi-dimensional spatial weight concern module is constructed, and the weights of different spatial information are preliminarily integrated to enhance the characteristics of rain lines. Then combining the advantages of dense connection and residual network, a dense residual fusion module is designed, which improves the learning ability of the network, realizes the reuse of information, and further correct the rain information. Finally, through the linear combination of various loss functions and rainy day imaging model, the output image quality is improved. Experimental results on several public data sets show that the subjective and objective evaluation indexes of the algorithm are better than those of the classical algorithm and novel algorithms, and the detailed information of the image background can be preserved more effectively while removing the rain pattern, which lays a foundation for the effective development of subsequent tasks based on machine vision.
In order to solve the problem that the poor image quality leads to the low efficiency of subsequent machine vision tasks in rainy days, an image rain removal algorithm based on multi-feature interaction and dense residual is proposed. Firstly, a multi-feature interactive convolution module is proposed to extract the semantic features of rain lines in different spaces in order to improve the utilization of information. Secondly, a multi-dimensional spatial weight concern module is constructed, and the weights of different spatial information are preliminarily integrated to enhance the characteristics of rain lines. Then combining the advantages of dense connection and residual network, a dense residual fusion module is designed, which improves the learning ability of the network, realizes the reuse of information, and further correct the rain information. Finally, through the linear combination of various loss functions and rainy day imaging model, the output image quality is improved. Experimental results on several public data sets show that the subjective and objective evaluation indexes of the algorithm are better than those of the classical algorithm and novel algorithms, and the detailed information of the image background can be preserved more effectively while removing the rain pattern, which lays a foundation for the effective development of subsequent tasks based on machine vision.
Abstract:
In order to study the finite-time guaranteed cost control problem for impulsive stochastic Markov jump systems with time-varying delays. By selecting mode-dependent Lyapunov functionals, using techniques such as linear matrix inequality (LMI) and average dwell time method, some new sufficient conditions are obtained to ensure finite-time stability of the system, and the upper bound performance index of the system is given. Based on the sufficient condition, the guaranteed cost controller with state feedback is designed. Finally, the LMI toolbox of matlab was used for data simulation, and the corresponding mean-square trajectory map was obtained. Numerical examples show that the obtained simulation results are consistent with the theoretical results, which verifies the effectiveness of the proposed theory.
In order to study the finite-time guaranteed cost control problem for impulsive stochastic Markov jump systems with time-varying delays. By selecting mode-dependent Lyapunov functionals, using techniques such as linear matrix inequality (LMI) and average dwell time method, some new sufficient conditions are obtained to ensure finite-time stability of the system, and the upper bound performance index of the system is given. Based on the sufficient condition, the guaranteed cost controller with state feedback is designed. Finally, the LMI toolbox of matlab was used for data simulation, and the corresponding mean-square trajectory map was obtained. Numerical examples show that the obtained simulation results are consistent with the theoretical results, which verifies the effectiveness of the proposed theory.
Abstract:
The rapid and accurate lane change path planning, effective tracking of the desired path, and maintaining vehicle stability during the lane change process are core technologies for ensuring active safety in intelligent vehicles. Aiming at the path planning problem in the process of active lane change of intelligent vehicles, the path planning strategy based on double quintic polynomials is proposed by introducing the intermediate position in this paper. It improves the smoothness of lane change path, ensures the safety of vehicle lane change, and satisfies the real-time requirement of lane change. The active lane change scenario is analyzed to determine the initial and target position of the lane change. The lane-changing path planning strategy of double quintic polynomial is proposed based on the critical collision point during lane change. Through the establishment of co-simulation model, the active lane change simulation test is carried out in different road states. The result shows that the lane-changing path planning by the double quintic polynomial programming method has more obvious lateral displacement due to the introduction of the intermediate point. And it can avoid the obstacle in front of the vehicle to ensure the safety of lane changing. At the same time, the maximum lateral acceleration of the vehicle at the intermediate position during lane changing is no more than 2m/s2, which ensures the handling stability of the vehicle during the lane change process. Under the conditions of dry or wet road surfaces, the required longitudinal safety distance for lane changing is reduced by about 20 meters, ensures the safety of longitudinal collisions during the lane changing process. The research results of the paper can provide theoretical and practical basis for the active lane changing path planning of intelligent vehicles.
The rapid and accurate lane change path planning, effective tracking of the desired path, and maintaining vehicle stability during the lane change process are core technologies for ensuring active safety in intelligent vehicles. Aiming at the path planning problem in the process of active lane change of intelligent vehicles, the path planning strategy based on double quintic polynomials is proposed by introducing the intermediate position in this paper. It improves the smoothness of lane change path, ensures the safety of vehicle lane change, and satisfies the real-time requirement of lane change. The active lane change scenario is analyzed to determine the initial and target position of the lane change. The lane-changing path planning strategy of double quintic polynomial is proposed based on the critical collision point during lane change. Through the establishment of co-simulation model, the active lane change simulation test is carried out in different road states. The result shows that the lane-changing path planning by the double quintic polynomial programming method has more obvious lateral displacement due to the introduction of the intermediate point. And it can avoid the obstacle in front of the vehicle to ensure the safety of lane changing. At the same time, the maximum lateral acceleration of the vehicle at the intermediate position during lane changing is no more than 2m/s2, which ensures the handling stability of the vehicle during the lane change process. Under the conditions of dry or wet road surfaces, the required longitudinal safety distance for lane changing is reduced by about 20 meters, ensures the safety of longitudinal collisions during the lane changing process. The research results of the paper can provide theoretical and practical basis for the active lane changing path planning of intelligent vehicles.
Abstract:
A three-stage rectifier containing three diodes is proposed in this paper. These diodes are located in different branches to achieve high conversion efficiency for different input power levels. By combining these branches, the proposed rectifier is able to achieve high conversion efficiency over a wide dynamic input power range. A prototype of a 2.4 GHz rectifier is designed, simulated and measured. The efficiency of each branch remains above 30% in the power range of -10 to 11 dBm, 11 to 24 dBm and 24 to 36 dBm, respectively. The measured results show that the conversion efficiency maintains over 40 % within the input power range of -6 to 25 dBm. The maximum conversion efficiency can reach 71.2%. It indicates that the proposed rectifier can achieve less sensitivity to input power variation and can achieve high conversion efficiency over an ultra-wide input power range, which can be used in energy harvesting or dynamic wireless power transfer systems.
A three-stage rectifier containing three diodes is proposed in this paper. These diodes are located in different branches to achieve high conversion efficiency for different input power levels. By combining these branches, the proposed rectifier is able to achieve high conversion efficiency over a wide dynamic input power range. A prototype of a 2.4 GHz rectifier is designed, simulated and measured. The efficiency of each branch remains above 30% in the power range of -10 to 11 dBm, 11 to 24 dBm and 24 to 36 dBm, respectively. The measured results show that the conversion efficiency maintains over 40 % within the input power range of -6 to 25 dBm. The maximum conversion efficiency can reach 71.2%. It indicates that the proposed rectifier can achieve less sensitivity to input power variation and can achieve high conversion efficiency over an ultra-wide input power range, which can be used in energy harvesting or dynamic wireless power transfer systems.
Abstract:
A reactive power optimization method for wind and solar power storage and distribution networks based on the Multi Objective Sand Cat swarm optimization (MOSCSO) is proposed to address the problems of slow convergence speed and easy falling into local optima in existing intelligent optimization algorithms for reactive power optimization in distribution networks. MOSCSO integrates the update and selection mechanism of external save sets in multi-objective algorithms, and has good global optimization ability. At the same time, the unique search and attack population update method of the sand cat swarm algorithm ensures its fast convergence speed and good optimization ability. Establish an IEEE 33 node system mathematical model with Energy Storage System (ESS) as the control variable, and apply MOSCSO for simulation verification. The results demonstrate that the proposed method in this paper can reduce grid losses and improve grid stability while balancing the wind and solar power generation system. Compared with traditional algorithms, the results verify the effectiveness and stability of MOSCSO in the reactive power optimization model of this paper.
A reactive power optimization method for wind and solar power storage and distribution networks based on the Multi Objective Sand Cat swarm optimization (MOSCSO) is proposed to address the problems of slow convergence speed and easy falling into local optima in existing intelligent optimization algorithms for reactive power optimization in distribution networks. MOSCSO integrates the update and selection mechanism of external save sets in multi-objective algorithms, and has good global optimization ability. At the same time, the unique search and attack population update method of the sand cat swarm algorithm ensures its fast convergence speed and good optimization ability. Establish an IEEE 33 node system mathematical model with Energy Storage System (ESS) as the control variable, and apply MOSCSO for simulation verification. The results demonstrate that the proposed method in this paper can reduce grid losses and improve grid stability while balancing the wind and solar power generation system. Compared with traditional algorithms, the results verify the effectiveness and stability of MOSCSO in the reactive power optimization model of this paper.
Analysis of temperature field of back-to-back Ω stator transverse flux permanent magnet linear motor
Abstract:
In consideration of the distinctive temperature field distribution of the back-to-back Ω stator transverse flux permanent magnet linear motor (BBΩ-TFPMLM), a "magneto-thermal coupling method" was employed to conduct in-depth research in this paper. Firstly, the finite element method (FEM) is used to model, analyze and optimize the BBΩ-TFPMLM structure, and determine the electromagnetic parameters in stable operation. At the same time, the magneto-thermal coupling model is built through Workbench and the real time monitoring temperature rise law of the loss model is obtained. Ultimately ,the temperature rises of the stator core and the rotor core under rated current are 29.21℃ and 11.12℃, each temperature rise is within the reasonable range. The thermal imager measurements of the temperature distribution at 750s are consistent with the simulation results, verifying the correctness of the proposed theory.
In consideration of the distinctive temperature field distribution of the back-to-back Ω stator transverse flux permanent magnet linear motor (BBΩ-TFPMLM), a "magneto-thermal coupling method" was employed to conduct in-depth research in this paper. Firstly, the finite element method (FEM) is used to model, analyze and optimize the BBΩ-TFPMLM structure, and determine the electromagnetic parameters in stable operation. At the same time, the magneto-thermal coupling model is built through Workbench and the real time monitoring temperature rise law of the loss model is obtained. Ultimately ,the temperature rises of the stator core and the rotor core under rated current are 29.21℃ and 11.12℃, each temperature rise is within the reasonable range. The thermal imager measurements of the temperature distribution at 750s are consistent with the simulation results, verifying the correctness of the proposed theory.
Abstract:
In the context of "dual-carbon", integrated energy system (IES) is of great significance to improve energy efficiency and reduce carbon emissions, in order to better achieve this goal, an IES low-carbon optimal dispatching method that considers hydrogen energy utilisation and demand response is proposed. On the source side, an IES model with hydrogen energy utilisation as the core is built to optimise the flexibility of equipment operation; on the load side, a demand response model based on a Logistic function is built to optimise the load curve and assist in carbon reduction, taking into account the energy consumption characteristics of users; at the same time, a stepped carbon trading mechanism is introduced into the optimisation model to further explore the carbon reduction potential of the system; finally, the system's energy purchase cost, operation and maintenance cost, carbon trading cost and abandoned wind power are taken into account. Finally, the system's energy purchase cost, operation and maintenance cost, carbon trading cost and wind abandonment cost are taken into account, and the total daily operating cost of IES is minimized as the objective function for optimal scheduling. The model not only achieves the effect of peak shaving and valley filling, but also effectively reduces the total operating cost and carbon emission of IES, which is a good low carbon economy.
In the context of "dual-carbon", integrated energy system (IES) is of great significance to improve energy efficiency and reduce carbon emissions, in order to better achieve this goal, an IES low-carbon optimal dispatching method that considers hydrogen energy utilisation and demand response is proposed. On the source side, an IES model with hydrogen energy utilisation as the core is built to optimise the flexibility of equipment operation; on the load side, a demand response model based on a Logistic function is built to optimise the load curve and assist in carbon reduction, taking into account the energy consumption characteristics of users; at the same time, a stepped carbon trading mechanism is introduced into the optimisation model to further explore the carbon reduction potential of the system; finally, the system's energy purchase cost, operation and maintenance cost, carbon trading cost and abandoned wind power are taken into account. Finally, the system's energy purchase cost, operation and maintenance cost, carbon trading cost and wind abandonment cost are taken into account, and the total daily operating cost of IES is minimized as the objective function for optimal scheduling. The model not only achieves the effect of peak shaving and valley filling, but also effectively reduces the total operating cost and carbon emission of IES, which is a good low carbon economy.
Abstract:
In order to address the high randomness and volatility of photovoltaic power, a single prediction model often struggles to accurately analyze the fluctuation patterns in historical data, resulting in low prediction accuracy. To overcome this issue, a combined model for short-term PV power prediction is proposed, which incorporates Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) and an improved Sparrow Search Algorithm (ISSA) for optimizing the Xtreme Gradient Boosting (XGBoost) model. First, remove the Outlier in the historical data and normalize them, and use Principal Component Analysis (PCA) to select features, so as to better identify the key factors affecting photovoltaic power. Then, the CNN network is used to extract the spatial features of the data, and the GRU network is used to extract the temporal features. In response to the difficulty in manually configuring parameters and high randomness of the XGBoost model, ISSA is used to optimize the hyperparameters of the model. Finally, the results predicted by the two methods are reduced using the reciprocal error method while updating the weights to obtain new predicted values to complete the prediction of photovoltaic power. The experimental results show that the proposed CNN-GRU-ISSA-XGBoost combination model has stronger adaptability and higher accuracy.
In order to address the high randomness and volatility of photovoltaic power, a single prediction model often struggles to accurately analyze the fluctuation patterns in historical data, resulting in low prediction accuracy. To overcome this issue, a combined model for short-term PV power prediction is proposed, which incorporates Convolutional Neural Network-Gated Recurrent Unit (CNN-GRU) and an improved Sparrow Search Algorithm (ISSA) for optimizing the Xtreme Gradient Boosting (XGBoost) model. First, remove the Outlier in the historical data and normalize them, and use Principal Component Analysis (PCA) to select features, so as to better identify the key factors affecting photovoltaic power. Then, the CNN network is used to extract the spatial features of the data, and the GRU network is used to extract the temporal features. In response to the difficulty in manually configuring parameters and high randomness of the XGBoost model, ISSA is used to optimize the hyperparameters of the model. Finally, the results predicted by the two methods are reduced using the reciprocal error method while updating the weights to obtain new predicted values to complete the prediction of photovoltaic power. The experimental results show that the proposed CNN-GRU-ISSA-XGBoost combination model has stronger adaptability and higher accuracy.
Abstract:
In order to ensure the output performance and suppress electromagnetic vibration of a hollow shaft permanent magnet brushless DC motor (HSPMBLDCM), a surrogate assisted model optimization method based on nonlinear multiple regression is proposed. Firstly, the optimal space filling sampling is determined by the Audze-Eglajs (AE) algorithm, and four main variables are selected by the Kernel Principal Components Analysis (KPCA) algorithm to construct the surrogate model. Next, the surrogate model was constructed using nonlinear multiple regression, and the R2 values of the coefficients of determination were all greater than 0.9, which verified the accuracy of the proxy model. Finally, the Robust Multiobjective Optimization Evolutionary Algorithms (RMOEA) were used to solve the surrogate model, and the optimal stator slot parameters were obtained. The results show that by optimizing the stator slot parameters, the average torque of the motor is reduced by 1.3%, which does not affect the output performance, and the maximum vibration acceleration of the motor at no load and rated load is reduced by 19% and 34.5%, respectively, which effectively reduces the electromagnetic vibration and proves the effectiveness and reliability of the optimization method. It provides an alternative means for motor vibration damping design optimization with outstanding practical significance.
In order to ensure the output performance and suppress electromagnetic vibration of a hollow shaft permanent magnet brushless DC motor (HSPMBLDCM), a surrogate assisted model optimization method based on nonlinear multiple regression is proposed. Firstly, the optimal space filling sampling is determined by the Audze-Eglajs (AE) algorithm, and four main variables are selected by the Kernel Principal Components Analysis (KPCA) algorithm to construct the surrogate model. Next, the surrogate model was constructed using nonlinear multiple regression, and the R2 values of the coefficients of determination were all greater than 0.9, which verified the accuracy of the proxy model. Finally, the Robust Multiobjective Optimization Evolutionary Algorithms (RMOEA) were used to solve the surrogate model, and the optimal stator slot parameters were obtained. The results show that by optimizing the stator slot parameters, the average torque of the motor is reduced by 1.3%, which does not affect the output performance, and the maximum vibration acceleration of the motor at no load and rated load is reduced by 19% and 34.5%, respectively, which effectively reduces the electromagnetic vibration and proves the effectiveness and reliability of the optimization method. It provides an alternative means for motor vibration damping design optimization with outstanding practical significance.
Abstract:
In order to solve the problem that the transmission tower-line system in landslide area is prone to rod failure under the action of slope deformation and wind load, a stress prediction model of tower-line system in landslide area based on IHPO-BP neural network is proposed. Firstly, the improved Circle chaotic mapping, nonlinear strategy of balance factor and Levy flight are used to optimize the hunter-prey optimization algorithm, and then the improved hunter-prey optimization algorithm is used to optimize the weights and thresholds of BP neural network, and the stress prediction model of tower-line system in landslide area is obtained. The wind direction angle, wind speed and displacement of tower leg support are used as model inputs, and the maximum stress of tower-line system members is used as output. The prediction results show that the IHPO-BP model proposed in this paper has high convergence speed and prediction accuracy. Compared with the HPO-BP model, the average absolute error decreases by 71.7%, the root mean square error decreases by 76.6% and the average relative error decreases by 71.8%.
In order to solve the problem that the transmission tower-line system in landslide area is prone to rod failure under the action of slope deformation and wind load, a stress prediction model of tower-line system in landslide area based on IHPO-BP neural network is proposed. Firstly, the improved Circle chaotic mapping, nonlinear strategy of balance factor and Levy flight are used to optimize the hunter-prey optimization algorithm, and then the improved hunter-prey optimization algorithm is used to optimize the weights and thresholds of BP neural network, and the stress prediction model of tower-line system in landslide area is obtained. The wind direction angle, wind speed and displacement of tower leg support are used as model inputs, and the maximum stress of tower-line system members is used as output. The prediction results show that the IHPO-BP model proposed in this paper has high convergence speed and prediction accuracy. Compared with the HPO-BP model, the average absolute error decreases by 71.7%, the root mean square error decreases by 76.6% and the average relative error decreases by 71.8%.
Abstract:
At present, the use of remote sensing technology in a timely and accurate monitoring of water quality, a comprehensive understanding of the distribution of water quality is of great significance to the protection of water environment.However, the relationship between water quality parameters and surface reflectance is not simple linear. BP neural network and SVM are widely used in water quality inversion because of their nonlinear simulation characteristics.However, the traditional BP neural network has the problems of slow convergence and easy to fall into local optimum. Although SVM has good fitting ability, it is greatly affected by penalty coefficient and kernel function parameter.Therefore, taking Yunlong Lake as the study area, using Sentinel-2 image and measured data, aiming at the important water quality parameters of conductivity and turbidity, a water quality inversion coupling model based on sparrow search algorithm (SSA) to optimize BP neural network and SVM was proposed. SSA was used to optimize the parameters of BP neural network and SVM, the model weight was calculated based on verification set MAE, and the final inversion result was obtained after the weighted calculation of output layer of SSA-BP and SSA-SVM model test group.And compared with BPNN, SVM, SSA-BP, SSA-SVM model.The results show that: (1) The sensitive bands of Sentinel-2 image to conductivity and turbidity are visible light and short wavelength infrared.(2) that precision of the couple model of SSA-BP-SVM is higher, and the R2 of the inversion model of conductivity and turbidity are 0.92 and 0.89 respectively.(3) the yunlong lake has the typical characteristics of urban water body, conductivity is greatly affected by the upstream of the south at water treatment plant drainage, turbidity is greatly affected by the social production activities of particulate pollutants.The SSA-BP-SVM model based on Sentinel-2 image has a good application potential in water quality inversion, which can provide a certain technical support for water quality monitoring and protection measures of Yunlong Lake.
At present, the use of remote sensing technology in a timely and accurate monitoring of water quality, a comprehensive understanding of the distribution of water quality is of great significance to the protection of water environment.However, the relationship between water quality parameters and surface reflectance is not simple linear. BP neural network and SVM are widely used in water quality inversion because of their nonlinear simulation characteristics.However, the traditional BP neural network has the problems of slow convergence and easy to fall into local optimum. Although SVM has good fitting ability, it is greatly affected by penalty coefficient and kernel function parameter.Therefore, taking Yunlong Lake as the study area, using Sentinel-2 image and measured data, aiming at the important water quality parameters of conductivity and turbidity, a water quality inversion coupling model based on sparrow search algorithm (SSA) to optimize BP neural network and SVM was proposed. SSA was used to optimize the parameters of BP neural network and SVM, the model weight was calculated based on verification set MAE, and the final inversion result was obtained after the weighted calculation of output layer of SSA-BP and SSA-SVM model test group.And compared with BPNN, SVM, SSA-BP, SSA-SVM model.The results show that: (1) The sensitive bands of Sentinel-2 image to conductivity and turbidity are visible light and short wavelength infrared.(2) that precision of the couple model of SSA-BP-SVM is higher, and the R2 of the inversion model of conductivity and turbidity are 0.92 and 0.89 respectively.(3) the yunlong lake has the typical characteristics of urban water body, conductivity is greatly affected by the upstream of the south at water treatment plant drainage, turbidity is greatly affected by the social production activities of particulate pollutants.The SSA-BP-SVM model based on Sentinel-2 image has a good application potential in water quality inversion, which can provide a certain technical support for water quality monitoring and protection measures of Yunlong Lake.
Abstract:
The holding of large-scale events will lead to the concentration of a large number of people and vehicles in the surrounding affected areas in a short period of time, and the road network around the venue is differentiated from conventional traffic. To investigate the mechanism of the impact of large events and their characteristics on the operation of the road network around the venues. The study resolved the influence characteristics of factors such as event scale, spatial distance between road sections and event venues, and constructed an interpretable machine learning model integrating XGBoost algorithm and Partial Dependence Plots to capture the nonlinear effects and synergistic influences of different factors. The study conducted an empirical study with Beijing as an example. The heterogeneity of the single factor shows that the spatial distance from the road section to the event venue and the event scale have a greater impact on the operational state of the road network around the venue, with a relative importance of 27.1% and 25.4%, respectively; The time from the start/end of the event has obvious non-linear characteristics on the road network operation status around the venue, and the road sections within 3 km around the venue will be significantly affected within 30-60 minutes before the event starts and 30 minutes after the event ends. The synergistic effect of two-dimensional factors shows that when the scale of the event is larger than 30,000 people, holidays and adverse weather have a negative impact on the running state of the road network around the venue. Meanwhile, it is found that in rain and haze weather, the operation status of the road network around the venue is affected by time and space, and the influence range is the road section within 2.5km from the venue within 60 minutes before the start and 40 minutes after the end of the event. The findings of the study can provide quantitative data support for identifying the causes of road congestion and formulating scientific and effective road network control strategies during large events.
The holding of large-scale events will lead to the concentration of a large number of people and vehicles in the surrounding affected areas in a short period of time, and the road network around the venue is differentiated from conventional traffic. To investigate the mechanism of the impact of large events and their characteristics on the operation of the road network around the venues. The study resolved the influence characteristics of factors such as event scale, spatial distance between road sections and event venues, and constructed an interpretable machine learning model integrating XGBoost algorithm and Partial Dependence Plots to capture the nonlinear effects and synergistic influences of different factors. The study conducted an empirical study with Beijing as an example. The heterogeneity of the single factor shows that the spatial distance from the road section to the event venue and the event scale have a greater impact on the operational state of the road network around the venue, with a relative importance of 27.1% and 25.4%, respectively; The time from the start/end of the event has obvious non-linear characteristics on the road network operation status around the venue, and the road sections within 3 km around the venue will be significantly affected within 30-60 minutes before the event starts and 30 minutes after the event ends. The synergistic effect of two-dimensional factors shows that when the scale of the event is larger than 30,000 people, holidays and adverse weather have a negative impact on the running state of the road network around the venue. Meanwhile, it is found that in rain and haze weather, the operation status of the road network around the venue is affected by time and space, and the influence range is the road section within 2.5km from the venue within 60 minutes before the start and 40 minutes after the end of the event. The findings of the study can provide quantitative data support for identifying the causes of road congestion and formulating scientific and effective road network control strategies during large events.
Abstract:
Not wearing helmets or wearing helmets incorrectly will pose significant threats to the safety of riders. Manual inspection not only requires large and inefficient work, but also makes it difficult to achieve full coverage of the entire area and time. This paper proposes a helmet wearing detection method for riders based on improved YOLOv5, which intelligently detects and automatically recognizes the helmet wearing status of riders through surveillance cameras. Firstly, a dataset of riders’ helmet wearing in different locations, perspectives, weather conditions, and time periods is created to lay the foundation for the research. Subsequently, a helmet wearing detection model based on improved YOLOv5 is proposed, which improves the multi-scale feature fusion module of YOLOv5 to enhance the detection effect of small targets. ECA attention mechanism is introduced to enhance feature map fusion effect, which significantly improves model detection accuracy; a lightweighted neck is designed based on GSConv to effectively reduce the detection time of the model. Our experiments show that the proposed algorithm has good detection performance for helmet wearing of riders, with mAP reaching 93.2%, which is 1.9% higher than the original algorithm. The detection time of a single image is 15.23ms, which ensures high detection rate and higher detection accuracy, and has certain application value.
Not wearing helmets or wearing helmets incorrectly will pose significant threats to the safety of riders. Manual inspection not only requires large and inefficient work, but also makes it difficult to achieve full coverage of the entire area and time. This paper proposes a helmet wearing detection method for riders based on improved YOLOv5, which intelligently detects and automatically recognizes the helmet wearing status of riders through surveillance cameras. Firstly, a dataset of riders’ helmet wearing in different locations, perspectives, weather conditions, and time periods is created to lay the foundation for the research. Subsequently, a helmet wearing detection model based on improved YOLOv5 is proposed, which improves the multi-scale feature fusion module of YOLOv5 to enhance the detection effect of small targets. ECA attention mechanism is introduced to enhance feature map fusion effect, which significantly improves model detection accuracy; a lightweighted neck is designed based on GSConv to effectively reduce the detection time of the model. Our experiments show that the proposed algorithm has good detection performance for helmet wearing of riders, with mAP reaching 93.2%, which is 1.9% higher than the original algorithm. The detection time of a single image is 15.23ms, which ensures high detection rate and higher detection accuracy, and has certain application value.
Abstract:
Southeast Asia is an important strategic position as an essential passage of China's "One Belt, One Road" strategy and an important node region for China's foreign trade and economic development. Based on NPP/VIIRS nighttime lighting data, the study uses spatial analysis methods such as center of gravity, standard deviation ellipse and Moran index to study the spatial and temporal characteristics of socio-economic development of Southeast Asian countries from 2012 to 2021. The results show that there is a significant correlation between GDP and nighttime light data of Southeast Asian countries; from the directional characteristics of Southeast Asian economic development, the center of gravity of Southeast Asian economy as a whole moves northward, the overall economic volume increases, the economic development gathers in the region, and the directionality of Southeast Asian economic development becomes more and more obvious; from the spatial aggregation characteristics of Southeast Asian economic development, high-high aggregation and low-low aggregation are the most In terms of the spatial aggregation characteristics of Southeast Asian economic development, high-high aggregation and low-low aggregation are the two most significant spatial aggregation characteristics. The high and high agglomeration areas play a good role in radiation and drive the economic development of the surrounding areas, while the low and high agglomeration areas have greater development potential. During the study time period, the low-low agglomeration areas in northern Southeast Asia decreased significantly.
Southeast Asia is an important strategic position as an essential passage of China's "One Belt, One Road" strategy and an important node region for China's foreign trade and economic development. Based on NPP/VIIRS nighttime lighting data, the study uses spatial analysis methods such as center of gravity, standard deviation ellipse and Moran index to study the spatial and temporal characteristics of socio-economic development of Southeast Asian countries from 2012 to 2021. The results show that there is a significant correlation between GDP and nighttime light data of Southeast Asian countries; from the directional characteristics of Southeast Asian economic development, the center of gravity of Southeast Asian economy as a whole moves northward, the overall economic volume increases, the economic development gathers in the region, and the directionality of Southeast Asian economic development becomes more and more obvious; from the spatial aggregation characteristics of Southeast Asian economic development, high-high aggregation and low-low aggregation are the most In terms of the spatial aggregation characteristics of Southeast Asian economic development, high-high aggregation and low-low aggregation are the two most significant spatial aggregation characteristics. The high and high agglomeration areas play a good role in radiation and drive the economic development of the surrounding areas, while the low and high agglomeration areas have greater development potential. During the study time period, the low-low agglomeration areas in northern Southeast Asia decreased significantly.
Abstract:
To address the problem of numerous hyperparameters, loss of long time series information and difficulty in distinguishing primary and secondary features in long short-term memory network (LSTM) for grain capacity prediction, so this paper proposed a data-driven grain capacity portfolio forecasting model. In the hyperparameter part, it performs hyperparameter search for LSTM by introducing dynamic weights and Laplace variants of the bald eagle algorithm (WLBES), avoiding the process of manual tuning of parameters; In the prediction part, it uses Ridge Regression (RR) to correct the residuals of the prediction results to make up for the deficiency of LSTM data loss; at the same time, it adds an attention mechanism to distinguish primary and secondary features by weight size to enhance the importance of features with greater relevance to food production. The results showed that the combined WLBES-LSTM-RR model decreased the root mean square error by 75% and 19% compared with the LSTM and WLBES-LSTM models, respectively, and a substantial decrease in RMSE compared with other combined models optimized for LSTM. This combined model has higher prediction accuracy in grain capacity prediction.
To address the problem of numerous hyperparameters, loss of long time series information and difficulty in distinguishing primary and secondary features in long short-term memory network (LSTM) for grain capacity prediction, so this paper proposed a data-driven grain capacity portfolio forecasting model. In the hyperparameter part, it performs hyperparameter search for LSTM by introducing dynamic weights and Laplace variants of the bald eagle algorithm (WLBES), avoiding the process of manual tuning of parameters; In the prediction part, it uses Ridge Regression (RR) to correct the residuals of the prediction results to make up for the deficiency of LSTM data loss; at the same time, it adds an attention mechanism to distinguish primary and secondary features by weight size to enhance the importance of features with greater relevance to food production. The results showed that the combined WLBES-LSTM-RR model decreased the root mean square error by 75% and 19% compared with the LSTM and WLBES-LSTM models, respectively, and a substantial decrease in RMSE compared with other combined models optimized for LSTM. This combined model has higher prediction accuracy in grain capacity prediction.
Abstract:
The air quality index (AQI) is a comprehensive indicator used to measure air quality conditions. The prediction of AQI can alert the public to air quality information and enable people to make more informed travel decisions. By predicting the change of air quality in advance, the government and environmental protection departments can take emergency measures to reduce air pollution. Therefore, this paper proposes an integrated deep learning model (CNN-GRU) based on convolutional neural network and gated recurrent unit for AQI prediction. In this paper, convolutional neural network (CNN) extracts the spatial and temporal characteristics of pollutant gas concentration and AQI and completes the feature mapping, while gated recurrent unit (GRU) models the temporal relationship and completes the calculation efficiently. The daily average concentrations and AQI of six major pollutant gases (PM2.5, PM10, SO2, CO, NO2, O3) in Beijing and Guangzhou from 2014-2022 are selected for example study, and the AQI is predicted using the CNN-GRU model, which is compared with the multiverse-optimized generalized regression neural network model (MVO-GRNN), genetic algorithm-optimized BP neural network model (GA-BP) for AQI prediction for comparative analysis. The experiments show that the CNN-GRU model proposed in this paper has the smallest prediction error for AQI.
The air quality index (AQI) is a comprehensive indicator used to measure air quality conditions. The prediction of AQI can alert the public to air quality information and enable people to make more informed travel decisions. By predicting the change of air quality in advance, the government and environmental protection departments can take emergency measures to reduce air pollution. Therefore, this paper proposes an integrated deep learning model (CNN-GRU) based on convolutional neural network and gated recurrent unit for AQI prediction. In this paper, convolutional neural network (CNN) extracts the spatial and temporal characteristics of pollutant gas concentration and AQI and completes the feature mapping, while gated recurrent unit (GRU) models the temporal relationship and completes the calculation efficiently. The daily average concentrations and AQI of six major pollutant gases (PM2.5, PM10, SO2, CO, NO2, O3) in Beijing and Guangzhou from 2014-2022 are selected for example study, and the AQI is predicted using the CNN-GRU model, which is compared with the multiverse-optimized generalized regression neural network model (MVO-GRNN), genetic algorithm-optimized BP neural network model (GA-BP) for AQI prediction for comparative analysis. The experiments show that the CNN-GRU model proposed in this paper has the smallest prediction error for AQI.
Abstract:
Accurate wind speed prediction is the key to large-scale application of wind energy in power systems, and the randomness and volatility of wind speed sequences make wind speed prediction more difficult. To enhance the predictability of wind speed sequences, a Logistic chaotic mapping strategy, adaptive parameter adjustment strategy, and the introduction of mutation strategy were used to improve the Carnivorous Plant Algorithm (CPA). A short-term wind speed prediction model based on error correction and VMD-ICPA-LSSVM was proposed. Firstly, meteorological factors are used as inputs for least squares support vector machine (LSSVM) to predict wind speed and obtain an error sequence. Then, K-L divergence is used to adaptively determine the parameters of Variational Mode Decomposition (VMD) and decompose the error sequence. Combining the Improved Carnivorous Plant Algorithm (ICPA) to optimize the adjustable parameters of LSSVM to predict the decomposed subsequences. After stacking the prediction results of each obtain the final wind speed prediction value. The experimental results show that subsequence, error correction is performed on the original prediction sequence to compared with other models, this model has better prediction accuracy and generalization performance.
Accurate wind speed prediction is the key to large-scale application of wind energy in power systems, and the randomness and volatility of wind speed sequences make wind speed prediction more difficult. To enhance the predictability of wind speed sequences, a Logistic chaotic mapping strategy, adaptive parameter adjustment strategy, and the introduction of mutation strategy were used to improve the Carnivorous Plant Algorithm (CPA). A short-term wind speed prediction model based on error correction and VMD-ICPA-LSSVM was proposed. Firstly, meteorological factors are used as inputs for least squares support vector machine (LSSVM) to predict wind speed and obtain an error sequence. Then, K-L divergence is used to adaptively determine the parameters of Variational Mode Decomposition (VMD) and decompose the error sequence. Combining the Improved Carnivorous Plant Algorithm (ICPA) to optimize the adjustable parameters of LSSVM to predict the decomposed subsequences. After stacking the prediction results of each obtain the final wind speed prediction value. The experimental results show that subsequence, error correction is performed on the original prediction sequence to compared with other models, this model has better prediction accuracy and generalization performance.
Abstract:
Aiming at the problems of boring training contents of traditional wrist rehabilitation training methods and low training efficiency due to low motivation of users to participate in training, a training system of myoelectric control virtual reality game was designed. The sEMG signals of wrist movement are collected and the wrist joint movement intention is decoded through the principle of muscle synergy for the control of the virtual reality game; random disturbance force is introduced in the virtual reality game, and the interaction with the virtual reality environment is realized through the way of impedance control, which makes users explore different movement control methods. The feasibility of the system was verified through model calibration experiments, and training experiments were conducted to assess the training effect by evaluating the task completion time as well as the path efficiency. The experimental results show that the task completion time was reduced by 58%; the path efficiency was improved by 49%, and the designed training system enables users to perform motion control in a more efficient way and improves the training efficiency.
Aiming at the problems of boring training contents of traditional wrist rehabilitation training methods and low training efficiency due to low motivation of users to participate in training, a training system of myoelectric control virtual reality game was designed. The sEMG signals of wrist movement are collected and the wrist joint movement intention is decoded through the principle of muscle synergy for the control of the virtual reality game; random disturbance force is introduced in the virtual reality game, and the interaction with the virtual reality environment is realized through the way of impedance control, which makes users explore different movement control methods. The feasibility of the system was verified through model calibration experiments, and training experiments were conducted to assess the training effect by evaluating the task completion time as well as the path efficiency. The experimental results show that the task completion time was reduced by 58%; the path efficiency was improved by 49%, and the designed training system enables users to perform motion control in a more efficient way and improves the training efficiency.
Abstract:
The multi-robot collaborative SLAM scheme solves the issues of high individual cost, global error accumulation, excessive concentration of calculation and risk under a single robot SLAM(Simultaneous Localization and Mapping) scheme, and has strong robustness and stability, which is a research hotspot when it comes to large-scale environmental mapping requirements. This study analyzes the history of this field's growth and introduces the multi-robot collaborative SLAM's fusion method and architecture. The current collaborative SLAM methods are arranged from the viewpoint of machine learning classification. The future development of deep learning, semantic maps, and multi-robot VSLAM, along with other significant multi-robot SLAM development directions, are also introduced at this time.
The multi-robot collaborative SLAM scheme solves the issues of high individual cost, global error accumulation, excessive concentration of calculation and risk under a single robot SLAM(Simultaneous Localization and Mapping) scheme, and has strong robustness and stability, which is a research hotspot when it comes to large-scale environmental mapping requirements. This study analyzes the history of this field's growth and introduces the multi-robot collaborative SLAM's fusion method and architecture. The current collaborative SLAM methods are arranged from the viewpoint of machine learning classification. The future development of deep learning, semantic maps, and multi-robot VSLAM, along with other significant multi-robot SLAM development directions, are also introduced at this time.
Abstract:
In view of the current problem that the railway safety zone division along the electrified railway with complex background needs to use actual fixed standard parts as reference and the division range is small, a smart zone division algorithm that does not require reference objects is proposed. This method first calculates the corresponding GSD parameters based on the relevant parameters in the images collected by the UAV, then uses DeepLabv3+ model with ECA-Net module to accurately segment the railway in the image. Then, a series of image processing operations such as edge detection, opening operation, and probability Hough transform are used to extract the key pixel points that make up the railway, and the least squares algorithm is used to fit the railway and obtain the mathematical expression of the railway. Finally, combined with mathematical algorithms, GSD parameters, and the mathematical expression of the railway, the division of the safety zone is completed. Experimental results show that the measurement accuracy of this method is over 90%, without the need to select fixed reference objects, and has strong adaptability and high robustness, which has high practicality and reliability and provides effective technical support for safety management along the electrified railway.
In view of the current problem that the railway safety zone division along the electrified railway with complex background needs to use actual fixed standard parts as reference and the division range is small, a smart zone division algorithm that does not require reference objects is proposed. This method first calculates the corresponding GSD parameters based on the relevant parameters in the images collected by the UAV, then uses DeepLabv3+ model with ECA-Net module to accurately segment the railway in the image. Then, a series of image processing operations such as edge detection, opening operation, and probability Hough transform are used to extract the key pixel points that make up the railway, and the least squares algorithm is used to fit the railway and obtain the mathematical expression of the railway. Finally, combined with mathematical algorithms, GSD parameters, and the mathematical expression of the railway, the division of the safety zone is completed. Experimental results show that the measurement accuracy of this method is over 90%, without the need to select fixed reference objects, and has strong adaptability and high robustness, which has high practicality and reliability and provides effective technical support for safety management along the electrified railway.
Abstract:
Illegal parking of vehicles will reduce the efficiency of road traffic, causing traffic congestion and traffic accidents. Traditional vehicle violation detection methods have a large number of parameters and low accuracy. Therefore, this paper proposes a vehicle violation detection method using the improved YOLOv5 model and radiographic method. Firstly, a lightweight feature extraction module is designed to reduce the amount of model parameters. Secondly, the attention mechanism is added to the model to enhance the feature extraction ability of the model from the channel dimension and spatial dimension to ensure the accuracy of the model. Then, the mixed data is used to enhance the samples of the rich dataset to improve the detection effect in complex backgrounds. Then, EIoU is selected as the loss function to improve the positioning ability of the model. Experiments show that the mean accuracy of the improved model reaches 91.35%, which is 1.01% higher than that of the original YOLOv5s, and the number of parameters is reduced by 35.79%. Finally, by combining the improved model with the ray method, the inspection speed on the Jetson Xavier NX embedded platform can reach about 28 frames per second, enabling real-time inspection.
Illegal parking of vehicles will reduce the efficiency of road traffic, causing traffic congestion and traffic accidents. Traditional vehicle violation detection methods have a large number of parameters and low accuracy. Therefore, this paper proposes a vehicle violation detection method using the improved YOLOv5 model and radiographic method. Firstly, a lightweight feature extraction module is designed to reduce the amount of model parameters. Secondly, the attention mechanism is added to the model to enhance the feature extraction ability of the model from the channel dimension and spatial dimension to ensure the accuracy of the model. Then, the mixed data is used to enhance the samples of the rich dataset to improve the detection effect in complex backgrounds. Then, EIoU is selected as the loss function to improve the positioning ability of the model. Experiments show that the mean accuracy of the improved model reaches 91.35%, which is 1.01% higher than that of the original YOLOv5s, and the number of parameters is reduced by 35.79%. Finally, by combining the improved model with the ray method, the inspection speed on the Jetson Xavier NX embedded platform can reach about 28 frames per second, enabling real-time inspection.
Abstract:
Two-dimensional (2D) materials have the characteristics of low loss, ultrafast carrier response, and wideband nonlinear saturable absorption properties, they have originated a range of innovative applications in photonics and photoelectric device owing to their advantages of a layered structure. Graphene-like materials have recently been utilized for short and ultrashort pulse laser generation in the visible, near-infrared, and mid-infrared wavelength regions. This article reviews the recent progress of 2D materials as saturable absorbers for Q-switched and mode-locked solid-state lasers. Firstly, the preparation methods of 2D materials as saturable absorbers, the saturable absorption principle, and methods for measuring nonlinear absorption properties are introduced and explained theoretically. Secondly, the solid-state pulsed laser is summarized based on the performance of 2D saturable absorbers in operating wavelength, output power, and pulse width. Finally, the development trend of two-dimensional saturable absorbers in solid-state lasers is suggested.
Two-dimensional (2D) materials have the characteristics of low loss, ultrafast carrier response, and wideband nonlinear saturable absorption properties, they have originated a range of innovative applications in photonics and photoelectric device owing to their advantages of a layered structure. Graphene-like materials have recently been utilized for short and ultrashort pulse laser generation in the visible, near-infrared, and mid-infrared wavelength regions. This article reviews the recent progress of 2D materials as saturable absorbers for Q-switched and mode-locked solid-state lasers. Firstly, the preparation methods of 2D materials as saturable absorbers, the saturable absorption principle, and methods for measuring nonlinear absorption properties are introduced and explained theoretically. Secondly, the solid-state pulsed laser is summarized based on the performance of 2D saturable absorbers in operating wavelength, output power, and pulse width. Finally, the development trend of two-dimensional saturable absorbers in solid-state lasers is suggested.
Abstract:
Sonar images are prone to problems such as low contrast, low resolution, and edge distortion, so it is difficult to accurately separate effective signals from noise when removing noise from sonar images, resulting in reduced image contrast, unclear edge contours, and severe detail loss after denoising. Therefore, this paper proposes a sonar image denoising algorithm based on adaptive Wiener filtering and 2D-VMD. Firstly, a noisy image is decomposed using two-dimensional variational mode decomposition to obtain a series of sub modes with different center frequencies. Effective modal components are filtered out using correlation coefficients and structural similarity, and use adaptive Wiener filtering to process effective modal components, and finally reconstruct the filtered modal components to remove noise.The experimental results show that the proposed image denoising algorithm achieves the best results in terms of correlation coefficient and structural similarity, with a peak signal-to-noise ratio slightly lower than that of NSST domain denoising. Taking into account objective data and visual effects, the algorithm proposed in this paper achieves the best results in image details and edge preservation after removing noise.
Sonar images are prone to problems such as low contrast, low resolution, and edge distortion, so it is difficult to accurately separate effective signals from noise when removing noise from sonar images, resulting in reduced image contrast, unclear edge contours, and severe detail loss after denoising. Therefore, this paper proposes a sonar image denoising algorithm based on adaptive Wiener filtering and 2D-VMD. Firstly, a noisy image is decomposed using two-dimensional variational mode decomposition to obtain a series of sub modes with different center frequencies. Effective modal components are filtered out using correlation coefficients and structural similarity, and use adaptive Wiener filtering to process effective modal components, and finally reconstruct the filtered modal components to remove noise.The experimental results show that the proposed image denoising algorithm achieves the best results in terms of correlation coefficient and structural similarity, with a peak signal-to-noise ratio slightly lower than that of NSST domain denoising. Taking into account objective data and visual effects, the algorithm proposed in this paper achieves the best results in image details and edge preservation after removing noise.
Abstract:
An algorithm based on improved YOLOv5s is proposed to address the problems of small percentage of traffic signs in the image, low detection accuracy and complex surrounding environment. Firstlythe attention mechanism ECA (Efficient Channel Attention) is added to the backbone network part to enhance the feature extraction ability of the network and effectively solve the problem of complex surrounding environment;Secondly, the HASPP (Hybrid Atrous Spatial Pyramid Pooling) is proposed, which enhances the network's ability to combine context; Finally,the Neck structure in the network is modified to allow efficient fusion of high level features with underlying features while avoiding information loss across convolutional layers. Experimental results show that the improved algorithm achieves an average detection accuracy of 94.4%, a recall rate of 74.1% and an accuracy rate of 94.0% on the traffic signage dataset, which were 3.6, 2.8, and 3.4 percentage points higher than the original algorithm, respectively.
An algorithm based on improved YOLOv5s is proposed to address the problems of small percentage of traffic signs in the image, low detection accuracy and complex surrounding environment. Firstlythe attention mechanism ECA (Efficient Channel Attention) is added to the backbone network part to enhance the feature extraction ability of the network and effectively solve the problem of complex surrounding environment;Secondly, the HASPP (Hybrid Atrous Spatial Pyramid Pooling) is proposed, which enhances the network's ability to combine context; Finally,the Neck structure in the network is modified to allow efficient fusion of high level features with underlying features while avoiding information loss across convolutional layers. Experimental results show that the improved algorithm achieves an average detection accuracy of 94.4%, a recall rate of 74.1% and an accuracy rate of 94.0% on the traffic signage dataset, which were 3.6, 2.8, and 3.4 percentage points higher than the original algorithm, respectively.
Abstract:
In order to solve the problem of reasonable and efficient route planning for takeout riders, an improved ant colony algorithm is proposed from the perspective of takeout riders. Firstly, the ant colony algorithm was used to obtain the initial planning path, and then the large-scale domain search algorithm was used to optimize the initial planning path, and the ant colony algorithm and large-scale domain search algorithm were combined to improve the solution quality. In order to verify the effectiveness of the method, the process of foreign distribution is simulated, and different order quantity scenarios are selected for comparative analysis. According to the optimal distribution roadmap and the optimal value of the target penalty function, it can be concluded that the improved ant colony algorithm is effective, which proves that the proposed improved ant colony algorithm can improve the delivery efficiency of takeout deliverers. The method proposed in this paper can not only improve the intelligent level of delivery, but also put forward a more humanized delivery method from the perspective of delivery riders, which supports the sustainable development of delivery system of Internet connected delivery platform.
In order to solve the problem of reasonable and efficient route planning for takeout riders, an improved ant colony algorithm is proposed from the perspective of takeout riders. Firstly, the ant colony algorithm was used to obtain the initial planning path, and then the large-scale domain search algorithm was used to optimize the initial planning path, and the ant colony algorithm and large-scale domain search algorithm were combined to improve the solution quality. In order to verify the effectiveness of the method, the process of foreign distribution is simulated, and different order quantity scenarios are selected for comparative analysis. According to the optimal distribution roadmap and the optimal value of the target penalty function, it can be concluded that the improved ant colony algorithm is effective, which proves that the proposed improved ant colony algorithm can improve the delivery efficiency of takeout deliverers. The method proposed in this paper can not only improve the intelligent level of delivery, but also put forward a more humanized delivery method from the perspective of delivery riders, which supports the sustainable development of delivery system of Internet connected delivery platform.
Abstract:
The track geometry state measuring instrument using GNSS positioning cannot work in GNSS rejection environment such as tunnel and subway. Since the railroad track is still close to the design alignment even if it is deformed, the difference between the actual track position and its design value always remains within a certain range. For this reason, a track irregularity detection algorithm considering design attitude assisted IMU/ODO was proposed by combining railroad design parameters with inertial guidance/odometer information, which combined design attitude with inertial guidance solved attitude for Kalman filtering and uses odometer velocity for heading projection. Through computational experiments, the effect of track design attitude information on the improvement of track irregularity detection accuracy was analyzed. The experimental results show that the railroad design attitude information can significantly improve the track irregularity detection accuracy, and the proposed method is comparable to the dynamic detection method based on the total station assisted by the 30m chord track irregularity detection accuracy, and the overall detection efficiency is high to meet the needs of daily track detection.
The track geometry state measuring instrument using GNSS positioning cannot work in GNSS rejection environment such as tunnel and subway. Since the railroad track is still close to the design alignment even if it is deformed, the difference between the actual track position and its design value always remains within a certain range. For this reason, a track irregularity detection algorithm considering design attitude assisted IMU/ODO was proposed by combining railroad design parameters with inertial guidance/odometer information, which combined design attitude with inertial guidance solved attitude for Kalman filtering and uses odometer velocity for heading projection. Through computational experiments, the effect of track design attitude information on the improvement of track irregularity detection accuracy was analyzed. The experimental results show that the railroad design attitude information can significantly improve the track irregularity detection accuracy, and the proposed method is comparable to the dynamic detection method based on the total station assisted by the 30m chord track irregularity detection accuracy, and the overall detection efficiency is high to meet the needs of daily track detection.
Abstract:
Global Navigation Satellite System (GNSS), Strapdown Inertial Navigation System (SINS) and visual sensors complement each other, and their information fusion can obtain high-precision, drift-free navigation and positioning information. Aiming at the problem that GNSS/SINS/vision fusion navigation is vulnerable to the impact of motion speed, light change, occlusion, etc., which leads to the decrease of navigation positioning accuracy and robustness, this paper adds the SoftLone robust kernel function to the cost function of the graph optimization framework, and sets the gross error test procedure of the measured value to reduce the negative impact of outliers. Further, the chi-square test is performed on the calculated residuals of the measured value, and the weight of the over-limit residual is reduced to improve the accuracy and robustness of the system. The experimental results show that the algorithm in this paper has higher accuracy and better robustness than the traditional algorithm without robust kernel function, outlier elimination strategy and chi-square test, and the algorithm with other robust kernel functions. It can greatly improve the positioning accuracy and robustness of GNSS/SINS/visual navigation. In large scale environment, there is no large drift errors, and absolute pose root mean square error is 0.735m, and the standard deviation of absolute pose error is 0.336m.
Global Navigation Satellite System (GNSS), Strapdown Inertial Navigation System (SINS) and visual sensors complement each other, and their information fusion can obtain high-precision, drift-free navigation and positioning information. Aiming at the problem that GNSS/SINS/vision fusion navigation is vulnerable to the impact of motion speed, light change, occlusion, etc., which leads to the decrease of navigation positioning accuracy and robustness, this paper adds the SoftLone robust kernel function to the cost function of the graph optimization framework, and sets the gross error test procedure of the measured value to reduce the negative impact of outliers. Further, the chi-square test is performed on the calculated residuals of the measured value, and the weight of the over-limit residual is reduced to improve the accuracy and robustness of the system. The experimental results show that the algorithm in this paper has higher accuracy and better robustness than the traditional algorithm without robust kernel function, outlier elimination strategy and chi-square test, and the algorithm with other robust kernel functions. It can greatly improve the positioning accuracy and robustness of GNSS/SINS/visual navigation. In large scale environment, there is no large drift errors, and absolute pose root mean square error is 0.735m, and the standard deviation of absolute pose error is 0.336m.
Abstract:
With the continuous intensification of global climate change and the rapid development of urbanization, urban waterlogging disasters caused by extreme rainfall processes have become increasingly severe and have become a serious challenge faced by many cities around the world. Based on the rainfall data from 75 national automatic meteorological observation stations in Zhuji City, Zhejiang Province from May to August 2021 and the depth data of typical waterlogging points, a deep learning model Long Short Term Memory (LSTM) network is used to construct a relationship model between rainfall and waterlogging depth, providing a 2-hour urban waterlogging level forecast with an interval of 15 minutes in the future, The prediction results are compared with those of random forest (RF) and Artificial Neural Network (ANN) models. The prediction results show that LSTM has the best performance in predicting water accumulation in the next 2 hours using the water accumulation and rainfall data from the first 4 hours, with a root mean square error (RMSE) of less than 5.6cm, a correlation coefficient (CC) of over 0.93, and a Nash efficiency coefficient (NSE) of over 0.86. The prediction effect is superior to RF and ANN, indicating that the constructed artificial intelligence model for water accumulation prediction has good prediction performance.
With the continuous intensification of global climate change and the rapid development of urbanization, urban waterlogging disasters caused by extreme rainfall processes have become increasingly severe and have become a serious challenge faced by many cities around the world. Based on the rainfall data from 75 national automatic meteorological observation stations in Zhuji City, Zhejiang Province from May to August 2021 and the depth data of typical waterlogging points, a deep learning model Long Short Term Memory (LSTM) network is used to construct a relationship model between rainfall and waterlogging depth, providing a 2-hour urban waterlogging level forecast with an interval of 15 minutes in the future, The prediction results are compared with those of random forest (RF) and Artificial Neural Network (ANN) models. The prediction results show that LSTM has the best performance in predicting water accumulation in the next 2 hours using the water accumulation and rainfall data from the first 4 hours, with a root mean square error (RMSE) of less than 5.6cm, a correlation coefficient (CC) of over 0.93, and a Nash efficiency coefficient (NSE) of over 0.86. The prediction effect is superior to RF and ANN, indicating that the constructed artificial intelligence model for water accumulation prediction has good prediction performance.
Abstract:
Reclaimed land is an important reserve land resource, but usually the soil structure is poor, and the organic matter and nutrient content are low. Increasing the application of organic fertilizers is a key way to quickly improve soil productivity, but it will cause large emissions of greenhouse gases such as nitrous oxide (N2O). Inoculation with plant rhizosphere growth promoters (PGPR) with N2O reduction function not only reduces greenhouse gas emissions, but also promotes crop growth. In this study, a PGPR denitrificans YSQ030 with N2O reduction function was used as the test strain to clarify the effect of inoculation YSQ030 on N2O emission and nitrogen cycling key functional genes in reclaimed soil with organic fertilizer application. By setting up soil microcosmic experiments for the application of commercial organic fertilizer and sheep manure organic fertilizer, the soil N2O emission flux after inoculation YSQ030 was analyzed by gas chromatography, and the cumulative emission was further calculated. Soil pH, EC, nitrate nitrogen and ammonium nitrogen content were analyzed at the end of the experiment, and the abundance of soil nitrification functional genes (AOA amoA and AOB amoA) and denitrification functional genes (nirS, nirK, nosZ I and nosZ II) was analyzed by real-time real-time quantitative PRC. The results showed that inoculation of YSQ030 in soil applied to commercial organic fertilizer and sheep manure organic fertilizer significantly reduced the N2O emission of reclaimed soil, and the maximum reduction of N2O emission was 90% and 30%, respectively. The N2O emissions of commercial organic fertilizer treatment were much higher than those of sheep manure organic fertilizer treatment, which may be due to the fact that the abundance of the N2O reductase gene nosZ I and the nosZ II gene of the N2O reductase gene of atypical denitrifying bacteria N2O reductase in the soil applied to commercial organic fertilizer was significantly lower than that of the soil where sheep manure organic fertilizer was applied. The application of commercial organic fertilizer significantly reduced the abundance of functional genes of soil nitrification and denitrification, while the application of sheep manure organic fertilizer had no obvious effect on the abundance of functional genes of soil nitrification and denitrification. This study shows that inoculation with YSQ030 can reduce the N2O emission in soil applied with organic fertilizer, which will provide a scientific basis for soil fertility improvement and N2O emission reduction, and will also provide core strain resources for the research and development of new microbial fertilizers or bio-organic fertilizers.
Reclaimed land is an important reserve land resource, but usually the soil structure is poor, and the organic matter and nutrient content are low. Increasing the application of organic fertilizers is a key way to quickly improve soil productivity, but it will cause large emissions of greenhouse gases such as nitrous oxide (N2O). Inoculation with plant rhizosphere growth promoters (PGPR) with N2O reduction function not only reduces greenhouse gas emissions, but also promotes crop growth. In this study, a PGPR denitrificans YSQ030 with N2O reduction function was used as the test strain to clarify the effect of inoculation YSQ030 on N2O emission and nitrogen cycling key functional genes in reclaimed soil with organic fertilizer application. By setting up soil microcosmic experiments for the application of commercial organic fertilizer and sheep manure organic fertilizer, the soil N2O emission flux after inoculation YSQ030 was analyzed by gas chromatography, and the cumulative emission was further calculated. Soil pH, EC, nitrate nitrogen and ammonium nitrogen content were analyzed at the end of the experiment, and the abundance of soil nitrification functional genes (AOA amoA and AOB amoA) and denitrification functional genes (nirS, nirK, nosZ I and nosZ II) was analyzed by real-time real-time quantitative PRC. The results showed that inoculation of YSQ030 in soil applied to commercial organic fertilizer and sheep manure organic fertilizer significantly reduced the N2O emission of reclaimed soil, and the maximum reduction of N2O emission was 90% and 30%, respectively. The N2O emissions of commercial organic fertilizer treatment were much higher than those of sheep manure organic fertilizer treatment, which may be due to the fact that the abundance of the N2O reductase gene nosZ I and the nosZ II gene of the N2O reductase gene of atypical denitrifying bacteria N2O reductase in the soil applied to commercial organic fertilizer was significantly lower than that of the soil where sheep manure organic fertilizer was applied. The application of commercial organic fertilizer significantly reduced the abundance of functional genes of soil nitrification and denitrification, while the application of sheep manure organic fertilizer had no obvious effect on the abundance of functional genes of soil nitrification and denitrification. This study shows that inoculation with YSQ030 can reduce the N2O emission in soil applied with organic fertilizer, which will provide a scientific basis for soil fertility improvement and N2O emission reduction, and will also provide core strain resources for the research and development of new microbial fertilizers or bio-organic fertilizers.
Abstract:
Tripping is a common fault in power transmission and distribution systems. In recent years, in order to deal with this kind of fault, academic circles have proposed protection methods based on relay protection action and electrical element action. However, these methods for electrical protection have hysteresis in handling tripping faults. Therefore, the prediction of tripping faults in advance plays a vital role in dealing with hidden problems and power recovery. In this paper, a method of power system trip fault prediction based on multi-source time series data is proposed. LSTM is used to extract the time characteristics of multivariate data, which alleviates the problem of RNN gradient disappearance on long time series. The model adds a peephole connection structure on the three-layer grid to enable a single unit to view the LSTM unit status in the previous stage, thereby strengthening the network timing memory capability. Secondly, we use L2 regularization measures such as parameter normalization to mitigate the impact of over fitting in fault prediction on the results. Finally, support vector machine classifier is introduced to improve the generalization ability and robustness of the overall model. The experimental data were obtained from relevant institutions of the State Grid of China. Experiments show that the proposed method has the advantage of high classification accuracy compared with existing data mining methods. In the last part of this paper, the practical application is discussed to prove its feasibility in the actual scene.
Tripping is a common fault in power transmission and distribution systems. In recent years, in order to deal with this kind of fault, academic circles have proposed protection methods based on relay protection action and electrical element action. However, these methods for electrical protection have hysteresis in handling tripping faults. Therefore, the prediction of tripping faults in advance plays a vital role in dealing with hidden problems and power recovery. In this paper, a method of power system trip fault prediction based on multi-source time series data is proposed. LSTM is used to extract the time characteristics of multivariate data, which alleviates the problem of RNN gradient disappearance on long time series. The model adds a peephole connection structure on the three-layer grid to enable a single unit to view the LSTM unit status in the previous stage, thereby strengthening the network timing memory capability. Secondly, we use L2 regularization measures such as parameter normalization to mitigate the impact of over fitting in fault prediction on the results. Finally, support vector machine classifier is introduced to improve the generalization ability and robustness of the overall model. The experimental data were obtained from relevant institutions of the State Grid of China. Experiments show that the proposed method has the advantage of high classification accuracy compared with existing data mining methods. In the last part of this paper, the practical application is discussed to prove its feasibility in the actual scene.
Abstract:
Abstract:Traditional drought monitoring indexes mainly consider a single factor and often cannot comprehensively reflect the drought situation. Based on remote sensing and land surface data Assimilation System (CLDAS) data of China Meteorological Administration, a daily scale integrated drought monitoring model was established by Gradient Boosting Machine (GBM), one of machine learning algorithm, with multiple influencing factors and drought index which can directly reflect drought degree as independent variables and comprehensive meteorological drought index (CI) as dependent variable. It was researched by taking Drought in North China from 2015 to 2018 as a case. The results show that the model monitoring results were significantly correlated with the CI calculated values of the observation stations. The coefficients of determination of the training and test sets were 0.945 and 0.655, respectively, and the root mean square error (RMSE) was 0.033 and 0.082, respectively. The integrated drought monitoring model had high accuracy. The consistency rate between the model monitoring and CI calculated values was above 65%, and the correlation coefficients with the standard precipitation evapotranspiration index (SPEI) and relative soil moisture(RSM) were 0.68 and 0.6, respectively, which could better reflect the meteorological drought and agricultural drought. Monitoring of typical drought condition shows that the integrated drought monitoring model can accurately identify the drought occurrence of drought , and represent the situation of comprehensive drought by considering various drought influencing factors.
Abstract:Traditional drought monitoring indexes mainly consider a single factor and often cannot comprehensively reflect the drought situation. Based on remote sensing and land surface data Assimilation System (CLDAS) data of China Meteorological Administration, a daily scale integrated drought monitoring model was established by Gradient Boosting Machine (GBM), one of machine learning algorithm, with multiple influencing factors and drought index which can directly reflect drought degree as independent variables and comprehensive meteorological drought index (CI) as dependent variable. It was researched by taking Drought in North China from 2015 to 2018 as a case. The results show that the model monitoring results were significantly correlated with the CI calculated values of the observation stations. The coefficients of determination of the training and test sets were 0.945 and 0.655, respectively, and the root mean square error (RMSE) was 0.033 and 0.082, respectively. The integrated drought monitoring model had high accuracy. The consistency rate between the model monitoring and CI calculated values was above 65%, and the correlation coefficients with the standard precipitation evapotranspiration index (SPEI) and relative soil moisture(RSM) were 0.68 and 0.6, respectively, which could better reflect the meteorological drought and agricultural drought. Monitoring of typical drought condition shows that the integrated drought monitoring model can accurately identify the drought occurrence of drought , and represent the situation of comprehensive drought by considering various drought influencing factors.
Abstract:
Aiming at the problems of long time-consuming and large memory consumption of A* algorithm in solving path trajectory, this paper proposes an improved A* algorithm based on adaptive step. Firstly, the priority order of the search direction is set according to the position relationship between the current point and the end point, reducing the redundant planning calculation on unreasonable directions. Secondly, the judgment condition for reaching the end point is modified to achieve path jumping during trajectory planning. Thirdly, an adaptive step size strategy is proposed to improve the efficiency of A* algorithm in trajectory planning. Finally, an eight-directional search method is proposed to address the issues of large memory usage and possible memory overflow when facing large maps. Experimental results show that compared with the original A* algorithm, the improved A* algorithm greatly improves the efficiency of trajectory planning, and the problem of large memory usage is also well solved.
Aiming at the problems of long time-consuming and large memory consumption of A* algorithm in solving path trajectory, this paper proposes an improved A* algorithm based on adaptive step. Firstly, the priority order of the search direction is set according to the position relationship between the current point and the end point, reducing the redundant planning calculation on unreasonable directions. Secondly, the judgment condition for reaching the end point is modified to achieve path jumping during trajectory planning. Thirdly, an adaptive step size strategy is proposed to improve the efficiency of A* algorithm in trajectory planning. Finally, an eight-directional search method is proposed to address the issues of large memory usage and possible memory overflow when facing large maps. Experimental results show that compared with the original A* algorithm, the improved A* algorithm greatly improves the efficiency of trajectory planning, and the problem of large memory usage is also well solved.
Abstract:
Flight test data processing is a key step in flight test. The flight test data is characterized by many types, large amount of data and complex algorithms. The flight test data processing system needs to have new capabilities such as rapid analysis of test data and timely sharing of cross-regional data. In this paper, on the basis of analyzing the original flight test data processing system, new methods such as data processing center and diversity point, diversity point and diversity point sharing, horizontal flight data splitting and separation processing are proposed, and a new flight test data processing system based on edge-cloud collaboration is designed. The test flight data processing, cross-regional data sharing and the whole process of subject implementation effect deduction and analysis are realized.
Flight test data processing is a key step in flight test. The flight test data is characterized by many types, large amount of data and complex algorithms. The flight test data processing system needs to have new capabilities such as rapid analysis of test data and timely sharing of cross-regional data. In this paper, on the basis of analyzing the original flight test data processing system, new methods such as data processing center and diversity point, diversity point and diversity point sharing, horizontal flight data splitting and separation processing are proposed, and a new flight test data processing system based on edge-cloud collaboration is designed. The test flight data processing, cross-regional data sharing and the whole process of subject implementation effect deduction and analysis are realized.
Abstract:
Aiming at the problems of timeliness and simplification of prediction functions in stock prediction models, this paper proposes a Bi-directional LSTM stock prediction model based on Long Short-Term Memory (LSTM) neural network, which combines self-attention mechanism and temporal convolution network (TCN). The learning unit and prediction unit in the BiLSTM-SA-TCN model can effectively learn important stock data, capture long-term dependency information, and output the predicted closing price of the next day's stock. The experimental results show that the prediction error RMSE of the BiLSTM-SA-TCN model is the smallest, the MAE is the smallest, and the fitting degree R2 is the best on multiple stock data sets. In the comparative experiments of different models, the BiLSTM-SA-TCN model has a higher generalization ability and a higher training efficiency.
Aiming at the problems of timeliness and simplification of prediction functions in stock prediction models, this paper proposes a Bi-directional LSTM stock prediction model based on Long Short-Term Memory (LSTM) neural network, which combines self-attention mechanism and temporal convolution network (TCN). The learning unit and prediction unit in the BiLSTM-SA-TCN model can effectively learn important stock data, capture long-term dependency information, and output the predicted closing price of the next day's stock. The experimental results show that the prediction error RMSE of the BiLSTM-SA-TCN model is the smallest, the MAE is the smallest, and the fitting degree R2 is the best on multiple stock data sets. In the comparative experiments of different models, the BiLSTM-SA-TCN model has a higher generalization ability and a higher training efficiency.
Abstract:
Scenic spot is one of the most characteristic nature reserves in China, and also the main carrier of tourism resources. Based on the POI (point of interest) data of national scenic spots and the statistical yearbook of urban construction, the study comprehensively uses ArcGIS spatial analysis and SPSS multiple regression statistical analysis and other methods to deeply explore the spatial distribution characteristics and main influencing factors of scenic tourism resources in cities above prefecture level in China, with a view to providing reference for the construction of the national park system, the construction of new urbanization and the optimal allocation of tourism resources in the future. The research finds that: ① China"s tourism resources present a macro pattern of strong in the southeast and weak in the northwest, showing a spatial structure of concentrated contiguous distribution with major urban agglomerations as the core, banded distribution of scenic spots above Grade A, and spot distribution of national scenic spots; ② The tourism resources have formed three low quality distribution belts, namely, "Shanxi Hubei Guangdong", "Inner Mongolia Qinghai Tibet" and the southeast coast. The "Inner Mongolia Qinghai Tibet" and the two low quality gathering belts along the southeast coast also show a strong construction advantage, which also reflects from the side that these two belt shaped gathering areas have great potential to improve the quality of the landscape. ③ Urban construction scale, urban population size, local government financial expenditure, the development of tertiary industry and urban economic development level are the main factors affecting the spatial distribution of urban tourism resources.
Scenic spot is one of the most characteristic nature reserves in China, and also the main carrier of tourism resources. Based on the POI (point of interest) data of national scenic spots and the statistical yearbook of urban construction, the study comprehensively uses ArcGIS spatial analysis and SPSS multiple regression statistical analysis and other methods to deeply explore the spatial distribution characteristics and main influencing factors of scenic tourism resources in cities above prefecture level in China, with a view to providing reference for the construction of the national park system, the construction of new urbanization and the optimal allocation of tourism resources in the future. The research finds that: ① China"s tourism resources present a macro pattern of strong in the southeast and weak in the northwest, showing a spatial structure of concentrated contiguous distribution with major urban agglomerations as the core, banded distribution of scenic spots above Grade A, and spot distribution of national scenic spots; ② The tourism resources have formed three low quality distribution belts, namely, "Shanxi Hubei Guangdong", "Inner Mongolia Qinghai Tibet" and the southeast coast. The "Inner Mongolia Qinghai Tibet" and the two low quality gathering belts along the southeast coast also show a strong construction advantage, which also reflects from the side that these two belt shaped gathering areas have great potential to improve the quality of the landscape. ③ Urban construction scale, urban population size, local government financial expenditure, the development of tertiary industry and urban economic development level are the main factors affecting the spatial distribution of urban tourism resources.
Abstract:
Aiming at increasing the accuracy and reliability of irradiance prediction, the paper proposes a prediction model of short-term irradiance based on improved Stacking ensemble learning and error correction. Firstly, through gradient boosting decision tree, the paper characteristically screens the original data set, removes redundant characteristics, and increases prediction accuracy and computing efficiency; then, an improved Stacking irradiance prediction model is established. In accordance with difference in prediction accuracy of different prediction models in the primary layer under K-fold cross-validation, the paper weights the prediction results, and applies Box-Cox to transform and process the training set data input from the first layer to second layer of Stacking, so as to increase the normality and homoscedasticity of prediction. Finally, the paper extracts the historical prediction error data, and applies random forest to construct an error model to further increase the prediction accuracy. Comparing with the traditional model and the classic Stacking model, the experimental results show that there are significantly improvements on the prediction performance of model.
Aiming at increasing the accuracy and reliability of irradiance prediction, the paper proposes a prediction model of short-term irradiance based on improved Stacking ensemble learning and error correction. Firstly, through gradient boosting decision tree, the paper characteristically screens the original data set, removes redundant characteristics, and increases prediction accuracy and computing efficiency; then, an improved Stacking irradiance prediction model is established. In accordance with difference in prediction accuracy of different prediction models in the primary layer under K-fold cross-validation, the paper weights the prediction results, and applies Box-Cox to transform and process the training set data input from the first layer to second layer of Stacking, so as to increase the normality and homoscedasticity of prediction. Finally, the paper extracts the historical prediction error data, and applies random forest to construct an error model to further increase the prediction accuracy. Comparing with the traditional model and the classic Stacking model, the experimental results show that there are significantly improvements on the prediction performance of model.
Abstract:
Pervasive edge computing allows peer devices to establish independent communication connections, which can help users process massive computing tasks with low delay. However, distributed devices cannot obtain the global system status of the network in real time, and cannot guarantee the fairness of resource utilization. To solve this problem, a resource allocation scheme for pervasive edge computing based on generative adversary network (GAN) is proposed. In this scheme, firstly, a multi-objective optimization problem is established based on minimizing the time delay and energy consumption, then the optimization problem is transformed into the maximum reward problem according to the random game theory, and then a computation offloading algorithm based on multi-agent imitation learning is proposed. The algorithm combines multi-agent generation against imitation learning (GAIL) and Markov strategy (MDP) to approximate the performance of experts, and realizes the online execution of the algorithm. Finally, combined with non-dominated sorting genetic algorithm II (NSGA-II) multi-objective optimization algorithm, the time delay and energy consumption are jointly optimized. Simulation results show that, compared with other edge computing resource allocation schemes, the time delay of the proposed solution is shortened by 30.8% and the energy consumption is reduced by 34.3%.
Pervasive edge computing allows peer devices to establish independent communication connections, which can help users process massive computing tasks with low delay. However, distributed devices cannot obtain the global system status of the network in real time, and cannot guarantee the fairness of resource utilization. To solve this problem, a resource allocation scheme for pervasive edge computing based on generative adversary network (GAN) is proposed. In this scheme, firstly, a multi-objective optimization problem is established based on minimizing the time delay and energy consumption, then the optimization problem is transformed into the maximum reward problem according to the random game theory, and then a computation offloading algorithm based on multi-agent imitation learning is proposed. The algorithm combines multi-agent generation against imitation learning (GAIL) and Markov strategy (MDP) to approximate the performance of experts, and realizes the online execution of the algorithm. Finally, combined with non-dominated sorting genetic algorithm II (NSGA-II) multi-objective optimization algorithm, the time delay and energy consumption are jointly optimized. Simulation results show that, compared with other edge computing resource allocation schemes, the time delay of the proposed solution is shortened by 30.8% and the energy consumption is reduced by 34.3%.
Construction of Ni2P/g-C3N4/ZnIn2S4 photocatalysts and their boosted photocatalytic reduction of CO2
Abstract:
Ternary composites Ni2P/g-C3N4/ZnIn2S4 were synthesized via a hydrothermal method, and their catalytic performance were evaluated by photoreduction of CO2. Kinds of characterizations (XRD, SEM, TEM, XPS, UV-vis, EIS, and PL) were applied to investigate the morphology, crystal structure, surface chemical states, band structure and photoelectric property of the composites. The results showed that the heterostructure with intense contact was constructed successfully via the facet engineering. Besides, the introduction of Ni2P and g-C3N4 could improve the band structure of photocatalysts, shorten the transmission distance of electrons and inhibit the recombination of photo-induced carriers effectively. Therefore, compared with pure g-C3N4 and binary composites g-C3N4/ZnIn2S4, ternary composites Ni2P/g-C3N4/ZnIn2S4 exhibited higher catalytic activity. Among Ni2P/g-C3N4/ZnIn2S4 composites, CNZ5 (Ni2P:g-C3N4:ZnIn2S4=1:5:7) revealed the optimal CO2 photoreduction efficiency, in which the yield amount of CH4, CH3OH, and HCOOH was 114.72 μmol·h?1·g?1, 17.38 μmol·h?1·g?1, and 20.15 μmol·h?1·g?1, respectively. In addition, the CO2 photoreduction mechanism was obtained by the application of in-situ DRIFTS, and the intermediates of HCO3- and HCOOH were found during the reaction process.
Ternary composites Ni2P/g-C3N4/ZnIn2S4 were synthesized via a hydrothermal method, and their catalytic performance were evaluated by photoreduction of CO2. Kinds of characterizations (XRD, SEM, TEM, XPS, UV-vis, EIS, and PL) were applied to investigate the morphology, crystal structure, surface chemical states, band structure and photoelectric property of the composites. The results showed that the heterostructure with intense contact was constructed successfully via the facet engineering. Besides, the introduction of Ni2P and g-C3N4 could improve the band structure of photocatalysts, shorten the transmission distance of electrons and inhibit the recombination of photo-induced carriers effectively. Therefore, compared with pure g-C3N4 and binary composites g-C3N4/ZnIn2S4, ternary composites Ni2P/g-C3N4/ZnIn2S4 exhibited higher catalytic activity. Among Ni2P/g-C3N4/ZnIn2S4 composites, CNZ5 (Ni2P:g-C3N4:ZnIn2S4=1:5:7) revealed the optimal CO2 photoreduction efficiency, in which the yield amount of CH4, CH3OH, and HCOOH was 114.72 μmol·h?1·g?1, 17.38 μmol·h?1·g?1, and 20.15 μmol·h?1·g?1, respectively. In addition, the CO2 photoreduction mechanism was obtained by the application of in-situ DRIFTS, and the intermediates of HCO3- and HCOOH were found during the reaction process.
Abstract:
In this paper, the authors used a simple hydrothermal method to prepare m-Bi2O4, and also used XRD, XPS, SEM, TEM, UV-vis DRS, PFM and other characterization methods to analyze the structure, morphology, surface valence and piezoelectric photocatalytic performance of the samples, and tested the piezoelectric photocatalytic activity of the material using sulfamethazine (SM) as a simulated pollutant. The results show that m-Bi2O4 has higher catalytic performance than BaTiO3 and BiOCl. After 60 min of the synergistic action of the piezoelectric, the degradation efficiency of the m-Bi2O4 to the SM is up to 96.46%. By changing the wavelength of light, it is confirmed that m-Bi2O4 still has high catalytic activity under the condition of light energy weakening. In addition, base on that capture experiment of active radicals, superoxide radicals with high oxidation activity and a small amount of hydroxyl radicals and singlet oxygen were produce in the reaction system. A possible mechanism of piezoelectric photocatalysis was proposed.
In this paper, the authors used a simple hydrothermal method to prepare m-Bi2O4, and also used XRD, XPS, SEM, TEM, UV-vis DRS, PFM and other characterization methods to analyze the structure, morphology, surface valence and piezoelectric photocatalytic performance of the samples, and tested the piezoelectric photocatalytic activity of the material using sulfamethazine (SM) as a simulated pollutant. The results show that m-Bi2O4 has higher catalytic performance than BaTiO3 and BiOCl. After 60 min of the synergistic action of the piezoelectric, the degradation efficiency of the m-Bi2O4 to the SM is up to 96.46%. By changing the wavelength of light, it is confirmed that m-Bi2O4 still has high catalytic activity under the condition of light energy weakening. In addition, base on that capture experiment of active radicals, superoxide radicals with high oxidation activity and a small amount of hydroxyl radicals and singlet oxygen were produce in the reaction system. A possible mechanism of piezoelectric photocatalysis was proposed.
Abstract:
In order to improve the problem of poor convergence performance and easy to fall into local optimum of SSA, the golden sine salp swarm algorithm (MGSSA) with multi-strategy is proposed. First, the selection opposition strategy is used to improve the population quality by calculating selection opposite solutions for individuals in the population that completely deviate from the optimal individual search direction. After that, the optimal individual and elite mean individual are added in the follower position update phase to speed up the convergence of the algorithm. Finally, the golden sine algorithm variation strategy is selected based on the probability to further improve the quality of the solution, while facilitating the algorithm to jump out of the local optimum later. In this study, experiments are conducted on 14 benchmark test functions to compare with other swarm intelligence optimization algorithms and the novel improved salp swarm algorithms, and its performance is applied to test the solution of engineering optimization problems in tension-pressure spring design problems. The results show that MGSSA has high convergence accuracy and stability, and performs well in solving engineering problems.
In order to improve the problem of poor convergence performance and easy to fall into local optimum of SSA, the golden sine salp swarm algorithm (MGSSA) with multi-strategy is proposed. First, the selection opposition strategy is used to improve the population quality by calculating selection opposite solutions for individuals in the population that completely deviate from the optimal individual search direction. After that, the optimal individual and elite mean individual are added in the follower position update phase to speed up the convergence of the algorithm. Finally, the golden sine algorithm variation strategy is selected based on the probability to further improve the quality of the solution, while facilitating the algorithm to jump out of the local optimum later. In this study, experiments are conducted on 14 benchmark test functions to compare with other swarm intelligence optimization algorithms and the novel improved salp swarm algorithms, and its performance is applied to test the solution of engineering optimization problems in tension-pressure spring design problems. The results show that MGSSA has high convergence accuracy and stability, and performs well in solving engineering problems.
Abstract:
In view of the multi-disciplinary integration characteristics of intelligent manufacturing engineering, the research and design of robot sorting experiment system based on 3D vision are carried out. Kinect camera, industrial robot, PC and end-effector are used to build the hardware experiment platform. A support vector machine (SVM) algorithm is designed to recognize target objects. Aiming at whether the objects to be recognized overlap and block each other, two strategies are designed to calculate the center position of the object based on Hough transform and pose estimation based on point cloud registration. A series of robot sorting experiments are completed under the guidance of the upper computer interactive interface. The experimental results show that the system can identify and sort the object of specific shape and color accurately and stably. The experiments involve the knowledge and technology of robot, machine learning, image processing, software and hardware design and many other courses. Because of its strong comprehensiveness and good openness,it provides a comprehensive and innovative practice platform for the construction of intelligent manufacturing engineering laboratory.
In view of the multi-disciplinary integration characteristics of intelligent manufacturing engineering, the research and design of robot sorting experiment system based on 3D vision are carried out. Kinect camera, industrial robot, PC and end-effector are used to build the hardware experiment platform. A support vector machine (SVM) algorithm is designed to recognize target objects. Aiming at whether the objects to be recognized overlap and block each other, two strategies are designed to calculate the center position of the object based on Hough transform and pose estimation based on point cloud registration. A series of robot sorting experiments are completed under the guidance of the upper computer interactive interface. The experimental results show that the system can identify and sort the object of specific shape and color accurately and stably. The experiments involve the knowledge and technology of robot, machine learning, image processing, software and hardware design and many other courses. Because of its strong comprehensiveness and good openness,it provides a comprehensive and innovative practice platform for the construction of intelligent manufacturing engineering laboratory.
Abstract:
When a vehicle turns at high speeds the body will tilt to the outside of the curve, which will lead to a rollover accident. Aiming at this problem, the active roll control of the vehicle body was studied to improve the steering stability of vehicle. A vehicle dynamic model with six degrees of freedom (DOFs), considering the yaw and roll motions, was established. The desired roll angle of the vehicle was determined and selected as the control objective, and a controller was designed to make the actual roll angle of the vehicle approach the desired roll angle. Under different driving conditions, vehicle body roll angle, dynamic deflection of the suspension, the acceleration perceived by occupants, the lateral load transfer rate, and the power consumption of active suspension for rolling control were investigated. Research results show that the active roll can make the actual roll angle rapidly approach to the desired roll angle, and still make the vehicle have a good driving stability under complex driving conditions; the active roll reduces the peak value of the suspension dynamic deflection, and makes the lateral acceleration perceived by occupants and the lateral load transfer rate quickly approach zero; the power consumption of the active suspension for rolling control is small, which ensures the vehicle has good economic performance.
When a vehicle turns at high speeds the body will tilt to the outside of the curve, which will lead to a rollover accident. Aiming at this problem, the active roll control of the vehicle body was studied to improve the steering stability of vehicle. A vehicle dynamic model with six degrees of freedom (DOFs), considering the yaw and roll motions, was established. The desired roll angle of the vehicle was determined and selected as the control objective, and a controller was designed to make the actual roll angle of the vehicle approach the desired roll angle. Under different driving conditions, vehicle body roll angle, dynamic deflection of the suspension, the acceleration perceived by occupants, the lateral load transfer rate, and the power consumption of active suspension for rolling control were investigated. Research results show that the active roll can make the actual roll angle rapidly approach to the desired roll angle, and still make the vehicle have a good driving stability under complex driving conditions; the active roll reduces the peak value of the suspension dynamic deflection, and makes the lateral acceleration perceived by occupants and the lateral load transfer rate quickly approach zero; the power consumption of the active suspension for rolling control is small, which ensures the vehicle has good economic performance.
Abstract:
In order to accurately analyze the dynamics and changing trends of KPI data in the daily monitoring of cloud computing clusters and predict the subsequent development to achieve the goal of improving the high availability of cloud computing clusters, this paper proposes a three-frequency cloud KPI data prediction based on the combined attention model EAAT method. First, low, medium and high frequency intrinsic mode variables (IMFs) of cloud KPI data are obtained based on Empirical wavelet transform (EWT) to reduce the complexity of data prediction. Secondly, according to the information characteristics of low, medium and high frequency IMFs obtained from the decomposition, the ARIMA, Autoformer, and TPA-BiLSTM models are used to predict each type of IMFs. Finally, the classification prediction results are combined through the inverse transformation IEWT to obtain the prediction result of the KPI. The prediction method in this paper has been verified on four data sets from Google and Amazon. Regardless of whether the data is periodic or not, the prediction method in this paper has good results. The root mean square error of the EWT-IF-LSTM model on the Google data set is reduced by about 11.26%.
In order to accurately analyze the dynamics and changing trends of KPI data in the daily monitoring of cloud computing clusters and predict the subsequent development to achieve the goal of improving the high availability of cloud computing clusters, this paper proposes a three-frequency cloud KPI data prediction based on the combined attention model EAAT method. First, low, medium and high frequency intrinsic mode variables (IMFs) of cloud KPI data are obtained based on Empirical wavelet transform (EWT) to reduce the complexity of data prediction. Secondly, according to the information characteristics of low, medium and high frequency IMFs obtained from the decomposition, the ARIMA, Autoformer, and TPA-BiLSTM models are used to predict each type of IMFs. Finally, the classification prediction results are combined through the inverse transformation IEWT to obtain the prediction result of the KPI. The prediction method in this paper has been verified on four data sets from Google and Amazon. Regardless of whether the data is periodic or not, the prediction method in this paper has good results. The root mean square error of the EWT-IF-LSTM model on the Google data set is reduced by about 11.26%.
Abstract:
Hydrogen peroxide (H2O2) is an environmentally friendly and efficient oxidant, which is widely used in medical, semiconductor chip and other industries. The electrochemical synthesis of hydrogen peroxide by oxygen reduction reaction (ORR) has great potential to replace the traditional anthraquinone method. To commercialize this process, the development of 2e- ORR electrocatalysts with high activity, high selectivity and long-term stability is imminently. Here, this article systematically presents the research of currently available metal and non-metal based catalysts, with special emphasis on the control strategy of surface groups, and resolves effects on bond binding strength and electron transfer pathways of intermediates in the reduction process. We focus on key strategies such as electronic and geometric effects, coordination heteroatom doping, and active sites of nonmetal-based materials, highlighting that appropriate meso-structural engineering and kinetic strategies can further optimize the catalytic activity and H2O2 selectivity of existing catalysts. Finally, the challenges of exploring the active centers of non-metallic catalysts, the influence of electrolyte environment on catalysts and industrial equipment design with larger output power are pointed out, and the directions of development in the field of electrocatalytic synthesis of hydrogen peroxide are presented.
Hydrogen peroxide (H2O2) is an environmentally friendly and efficient oxidant, which is widely used in medical, semiconductor chip and other industries. The electrochemical synthesis of hydrogen peroxide by oxygen reduction reaction (ORR) has great potential to replace the traditional anthraquinone method. To commercialize this process, the development of 2e- ORR electrocatalysts with high activity, high selectivity and long-term stability is imminently. Here, this article systematically presents the research of currently available metal and non-metal based catalysts, with special emphasis on the control strategy of surface groups, and resolves effects on bond binding strength and electron transfer pathways of intermediates in the reduction process. We focus on key strategies such as electronic and geometric effects, coordination heteroatom doping, and active sites of nonmetal-based materials, highlighting that appropriate meso-structural engineering and kinetic strategies can further optimize the catalytic activity and H2O2 selectivity of existing catalysts. Finally, the challenges of exploring the active centers of non-metallic catalysts, the influence of electrolyte environment on catalysts and industrial equipment design with larger output power are pointed out, and the directions of development in the field of electrocatalytic synthesis of hydrogen peroxide are presented.
Abstract:
Aiming at the distributed convex optimization problem in multi-agent systems, a zero-gradient-sum optimization algorithm based on an adaptive event-triggered mechanism is proposed in this paper. An adaptive event-triggered condition is designed based on the virtual clock, and the condition is triggered only when the virtual clock of each agent meets the condition, which effectively reduces the update times of the controller and the communication burden of the system. It is proved that the states of all agents converge asymptotically to the global optimal solution under the algorithm by constructing the Lyapunov function. In addition, the designed event-triggered condition makes the minimum inter-event time designable, effectively excluding Zeno behavior. Finally, the simulation results verify the effectiveness of the algorithm.
Aiming at the distributed convex optimization problem in multi-agent systems, a zero-gradient-sum optimization algorithm based on an adaptive event-triggered mechanism is proposed in this paper. An adaptive event-triggered condition is designed based on the virtual clock, and the condition is triggered only when the virtual clock of each agent meets the condition, which effectively reduces the update times of the controller and the communication burden of the system. It is proved that the states of all agents converge asymptotically to the global optimal solution under the algorithm by constructing the Lyapunov function. In addition, the designed event-triggered condition makes the minimum inter-event time designable, effectively excluding Zeno behavior. Finally, the simulation results verify the effectiveness of the algorithm.
Abstract:
In order to effectively guide the formulation of agricultural greenhouse gas emission reduction policies, CH4MOD model and emission factor method were used to estimate the emissions of methane (CH4) and nitrous oxide (N2O) from crops in the Yangtze River Delta (YRD) in 2018, and their temporal and spatial distribution were further analyzed, Finally, 1km × 1km grid greenhouse gas emission inventory of crops in the Yangtze River Delta in 2018 was established. The results showed that, the CH4 emission factor of paddy field in the YRD was 348.54 kg/hm2 and the nitrous oxide (N2O) emission factor of crops was 0.95 kg/hm2, which were consistent with previous research results. In 2018, 1.769 million tons of CH4 were discharged from paddy fields in the YRD (equivalent to 37.149 million tons of CO2), mainly from the Single-season paddy fields. 15114.9 tons of N2O were discharged from crops in the YRD in 2018 (equivalent to 4.504 million tons of CO2), and wheat was the largest contributor. The CH4 and N2O emissions were mainly originated from Jiangsu and Anhui provinces, especially from April to August. It is suggested to strengthen the control of agricultural greenhouse gas emissions and reduce the input of farmland carbon and nitrogen from the source.
In order to effectively guide the formulation of agricultural greenhouse gas emission reduction policies, CH4MOD model and emission factor method were used to estimate the emissions of methane (CH4) and nitrous oxide (N2O) from crops in the Yangtze River Delta (YRD) in 2018, and their temporal and spatial distribution were further analyzed, Finally, 1km × 1km grid greenhouse gas emission inventory of crops in the Yangtze River Delta in 2018 was established. The results showed that, the CH4 emission factor of paddy field in the YRD was 348.54 kg/hm2 and the nitrous oxide (N2O) emission factor of crops was 0.95 kg/hm2, which were consistent with previous research results. In 2018, 1.769 million tons of CH4 were discharged from paddy fields in the YRD (equivalent to 37.149 million tons of CO2), mainly from the Single-season paddy fields. 15114.9 tons of N2O were discharged from crops in the YRD in 2018 (equivalent to 4.504 million tons of CO2), and wheat was the largest contributor. The CH4 and N2O emissions were mainly originated from Jiangsu and Anhui provinces, especially from April to August. It is suggested to strengthen the control of agricultural greenhouse gas emissions and reduce the input of farmland carbon and nitrogen from the source.
Abstract:
Accurate prediction of wind speed in extreme weather can provide important guidance for disaster prevention and resistance of the distribution network. This paper proposes a method based on temporal convolutional network (TCN) and Bi-directional Long Short Term Memory（BiLSTM）and error correction model for wind speed prediction in extreme weather. Firstly, the weather data is preprocessed. The time series characteristics of multi-feature data are extracted by TCN, and the extracted information is input into BiLSTM for wind speed prediction. In order to further improve the prediction accuracy, variational mode decomposition (VMD) is introduced to decompose the error sequence, and constructing BiLSTM models for the decomposed error subsequences respectively to perform error prediction. Then the error prediction value is used to correct the wind speed prediction value. Combined with the measured weather data in a certain place in Henan Province, the simulation results verified that the proposed method can effectively predict wind speed and has a high prediction accuracy when extreme weather occurs.
Accurate prediction of wind speed in extreme weather can provide important guidance for disaster prevention and resistance of the distribution network. This paper proposes a method based on temporal convolutional network (TCN) and Bi-directional Long Short Term Memory（BiLSTM）and error correction model for wind speed prediction in extreme weather. Firstly, the weather data is preprocessed. The time series characteristics of multi-feature data are extracted by TCN, and the extracted information is input into BiLSTM for wind speed prediction. In order to further improve the prediction accuracy, variational mode decomposition (VMD) is introduced to decompose the error sequence, and constructing BiLSTM models for the decomposed error subsequences respectively to perform error prediction. Then the error prediction value is used to correct the wind speed prediction value. Combined with the measured weather data in a certain place in Henan Province, the simulation results verified that the proposed method can effectively predict wind speed and has a high prediction accuracy when extreme weather occurs.
Abstract:
Micro-expression is the facial expression that people reveal involuntarily when they try to hide their true emotions. It is a hot research field in the field of affective computing in recent years. However, micro-expression recognition is a challenging task due to its short duration and low intensity. Based on the excellent performance of CrossViT in the field of image classification, this paper uses CrossViT as the backbone network to improve the cross-attention mechanism in the network. DA module (Dual Attention) is proposed to extend the traditional cross-attention mechanism and determine the correlation between attention results. Thus, the precision of micro-expression recognition is improved. The network learns from three optical flow features (optical strain, horizontal and vertical optical flow fields), which are calculated from the starting frame and peak frame of each micro-expression sequence. Finally, the micro-expression is classified by Softmax. In the microexpression fusion data set, UF1 and UAR reach 0.7275 and 0.7272, respectively, and the recognition accuracy is better than the mainstream algorithms in the field of microexpression, indicating the effectiveness of the network proposed in this paper.
Micro-expression is the facial expression that people reveal involuntarily when they try to hide their true emotions. It is a hot research field in the field of affective computing in recent years. However, micro-expression recognition is a challenging task due to its short duration and low intensity. Based on the excellent performance of CrossViT in the field of image classification, this paper uses CrossViT as the backbone network to improve the cross-attention mechanism in the network. DA module (Dual Attention) is proposed to extend the traditional cross-attention mechanism and determine the correlation between attention results. Thus, the precision of micro-expression recognition is improved. The network learns from three optical flow features (optical strain, horizontal and vertical optical flow fields), which are calculated from the starting frame and peak frame of each micro-expression sequence. Finally, the micro-expression is classified by Softmax. In the microexpression fusion data set, UF1 and UAR reach 0.7275 and 0.7272, respectively, and the recognition accuracy is better than the mainstream algorithms in the field of microexpression, indicating the effectiveness of the network proposed in this paper.
Abstract:
Aiming at the problem of disorderly charging and discharging of cell access charging piles to increase the load peak-to-valley difference rate and user cost, an energy storage charging pile orderly charging and discharging optimization operation strategy was proposed. Under the premise of reducing the peak-to-valley difference rate, the optimization strategy aims to optimize the lowest charging cost and the highest profit of charging pile. A typical day is selected to establish the optimal charge and discharge scheduling model of the charging pile, which is solved by improving the multi-objective particle swarm optimization algorithm, and the charge and discharge power of the energy storage charging pile is adjusted in combination with the time-of-use electricity price. Experimental data show that the improved method can effectively improve the convergence speed of the algorithm and better deal with multi-objective problems, the dispatching method can effectively reduce the peak-to-valley difference rate of typical load by 55%, 36% better than the original algorithm, reasonably allocate charging piles to store power resources during valley hours, effectively reduce user charging costs by 20%-30%, and improve charging pile profits, so as to achieve the win-win goal of power grid, user and charging pile.
Aiming at the problem of disorderly charging and discharging of cell access charging piles to increase the load peak-to-valley difference rate and user cost, an energy storage charging pile orderly charging and discharging optimization operation strategy was proposed. Under the premise of reducing the peak-to-valley difference rate, the optimization strategy aims to optimize the lowest charging cost and the highest profit of charging pile. A typical day is selected to establish the optimal charge and discharge scheduling model of the charging pile, which is solved by improving the multi-objective particle swarm optimization algorithm, and the charge and discharge power of the energy storage charging pile is adjusted in combination with the time-of-use electricity price. Experimental data show that the improved method can effectively improve the convergence speed of the algorithm and better deal with multi-objective problems, the dispatching method can effectively reduce the peak-to-valley difference rate of typical load by 55%, 36% better than the original algorithm, reasonably allocate charging piles to store power resources during valley hours, effectively reduce user charging costs by 20%-30%, and improve charging pile profits, so as to achieve the win-win goal of power grid, user and charging pile.
Abstract:
It has become a mandatory requirement of the traffic management department that electric vehicles must wear safety helmets on the road. In order to automatically detect the wearing of helmets by electric vehicle riders, a helmet and license plate detection method based on the improved YOLOv5m model is proposed, which can detect the electric vehicle license plate while detecting that the riders do not wear helmets, so as to track down the violators. The model is trained with self built data set, and DIOU loss function is used to replace GIOU loss function, DIOU_NMS replaces weighted NMS to enhance the recognition ability of the model for dense cycling scenes. At the same time, ECA attention mechanism is added to the Backone part and the Neck part for predicting small and medium-sized targets, which improves the recognition rate of the model for small and medium-sized targets. Then, the Kmeans algorithm is used to re-cluster the anchor frame size. Finally, the Mosaic data enhancement method is improved.The experimental results show that the map of the improved YOLOv5m electric vehicle rider helmet and license plate detection model is 92.7%, which is 2.15% higher than the original YOLOv5m model, and 5.7% and 6.9% higher than the YOLOv4 tiny and Faster RCNN models respectively. The improved YOLOv5m model can effectively improve the recognition rate of helmet and license plate.
It has become a mandatory requirement of the traffic management department that electric vehicles must wear safety helmets on the road. In order to automatically detect the wearing of helmets by electric vehicle riders, a helmet and license plate detection method based on the improved YOLOv5m model is proposed, which can detect the electric vehicle license plate while detecting that the riders do not wear helmets, so as to track down the violators. The model is trained with self built data set, and DIOU loss function is used to replace GIOU loss function, DIOU_NMS replaces weighted NMS to enhance the recognition ability of the model for dense cycling scenes. At the same time, ECA attention mechanism is added to the Backone part and the Neck part for predicting small and medium-sized targets, which improves the recognition rate of the model for small and medium-sized targets. Then, the Kmeans algorithm is used to re-cluster the anchor frame size. Finally, the Mosaic data enhancement method is improved.The experimental results show that the map of the improved YOLOv5m electric vehicle rider helmet and license plate detection model is 92.7%, which is 2.15% higher than the original YOLOv5m model, and 5.7% and 6.9% higher than the YOLOv4 tiny and Faster RCNN models respectively. The improved YOLOv5m model can effectively improve the recognition rate of helmet and license plate.
Abstract:
The factors that affect stock price are very complicated. Therefore, this paper revises the LSTM, which is commonly used in time series, and selects stock price for prediction under the condition of multivariable. First, variance inflation factor（VIF） was used to screen variables, and then the adaptive promotion method (Adaboost) model is combined to check the importance of characteristic variables. Secondly, the crawler is used to conduct text analysis of investor sentiment, calculate sentiment index and other indicators, and reveal the relationship between them and stock prices. Then, three different stock prices of Gree Electric Appliances, Flyco Electric Appliances and Midea Group were predicted, multi-layer perceptron（MPL）model and LSTM model were compared, and the appropriate model was selected as the benchmark model. On the basis of the benchmark model, the LSTM-EM model was constructed by adding sentiment index, investor attention and other indicators. Furthermore, GM (1,1) model was used to correct the residual term after considering investor sentiment. The empirical results show that the model can predict the stock price accurately.
The factors that affect stock price are very complicated. Therefore, this paper revises the LSTM, which is commonly used in time series, and selects stock price for prediction under the condition of multivariable. First, variance inflation factor（VIF） was used to screen variables, and then the adaptive promotion method (Adaboost) model is combined to check the importance of characteristic variables. Secondly, the crawler is used to conduct text analysis of investor sentiment, calculate sentiment index and other indicators, and reveal the relationship between them and stock prices. Then, three different stock prices of Gree Electric Appliances, Flyco Electric Appliances and Midea Group were predicted, multi-layer perceptron（MPL）model and LSTM model were compared, and the appropriate model was selected as the benchmark model. On the basis of the benchmark model, the LSTM-EM model was constructed by adding sentiment index, investor attention and other indicators. Furthermore, GM (1,1) model was used to correct the residual term after considering investor sentiment. The empirical results show that the model can predict the stock price accurately.
Abstract:
In order to improve the accuracy of short-term wind direction forecasting, a hybrid model, named EEMD-CNN-GRU, is proposed based on ensemble empirical mode decomposition (EEMD), convolutional neural network (CNN) and gated recurrent unit (GRU). According to the characteristics of randomness and unsteadiness of wind direction series, EEMD is used to decompose the data into multiple components. Then, CNN’s local connection and weight sharing are harnessed to extract the potential features in the components. Finally, GRU is adopted to further construct features of the potential features extracted by CNN, and the predicted values of each component are superposed to obtain the ultimate prediction results. The experimental results show that the proposed prediction method achieves good performance compared with other models such as BP neural network and long short-term memory (LSTM).
In order to improve the accuracy of short-term wind direction forecasting, a hybrid model, named EEMD-CNN-GRU, is proposed based on ensemble empirical mode decomposition (EEMD), convolutional neural network (CNN) and gated recurrent unit (GRU). According to the characteristics of randomness and unsteadiness of wind direction series, EEMD is used to decompose the data into multiple components. Then, CNN’s local connection and weight sharing are harnessed to extract the potential features in the components. Finally, GRU is adopted to further construct features of the potential features extracted by CNN, and the predicted values of each component are superposed to obtain the ultimate prediction results. The experimental results show that the proposed prediction method achieves good performance compared with other models such as BP neural network and long short-term memory (LSTM).
Abstract:
In view of the problem that the traditional static parameter equivalent circuit model of the supercapacitor cannot effectively reflect its dynamic operating characteristics, a second-order ladder equivalent circuit model including dynamic charge/discharge resistance and capacitance parameters is proposed. The recursive least square method is used for the offline parameter identification of the second-order ladder equivalent circuit model at first. Furthermore, by considering the dynamic change of the supercapacitor parameters, the offline identified model parameters are used as initial values. Then, the recursive least square method with forgetting factor is introduced to identify the dynamic parameters of supercapacitor in real-time. Simulation model and experimental test platform of the supercapacitor are established. On this basis, the effectiveness and accuracy of the second-order ladder equivalent circuit model with dynamic parameters and the real-time identification method are verified through the comparison of simulation and experimental results. Results show that the second-order ladder equivalent circuit model with dynamic parameters can effectively reflect the dynamic charge/discharge operating characteristics of the supercapacitor. Compared with the traditional three branch and second-order ladder equivalent circuit models with static parameters, the model accuracy is improved over 2.08% and 3.56%, respectively.
In view of the problem that the traditional static parameter equivalent circuit model of the supercapacitor cannot effectively reflect its dynamic operating characteristics, a second-order ladder equivalent circuit model including dynamic charge/discharge resistance and capacitance parameters is proposed. The recursive least square method is used for the offline parameter identification of the second-order ladder equivalent circuit model at first. Furthermore, by considering the dynamic change of the supercapacitor parameters, the offline identified model parameters are used as initial values. Then, the recursive least square method with forgetting factor is introduced to identify the dynamic parameters of supercapacitor in real-time. Simulation model and experimental test platform of the supercapacitor are established. On this basis, the effectiveness and accuracy of the second-order ladder equivalent circuit model with dynamic parameters and the real-time identification method are verified through the comparison of simulation and experimental results. Results show that the second-order ladder equivalent circuit model with dynamic parameters can effectively reflect the dynamic charge/discharge operating characteristics of the supercapacitor. Compared with the traditional three branch and second-order ladder equivalent circuit models with static parameters, the model accuracy is improved over 2.08% and 3.56%, respectively.
Abstract:
In order to reduce the high deviation between the predicted value and the observed value of the traditional numerical prediction model (GRAPES_GFS) of 2m temperature and improve the prediction accuracy. Combining the grid data of GRAPES_GFS and the corresponding observation data, this paper proposes a 2m temperature forecast correction model based on Convolution Long Short Time Memory Network (ConvLSTM) and attention mechanism. The model mainly includes the following steps: First, the local shallow features of the input data are extracted by the local feature extraction module. Secondly, input the extracted feature map into the feature attention module, assign different weights to different channel dimensions and different spatial dimensions of the data, restrain the weight of meteorological elements with low correlation with 2m temperature, and achieve local enhancement of high-temperature areas in 2m temperature data. Finally, ConvLSTM network is used to capture the time dimension characteristics of data and output the forecast correction results. The experimental results show that, compared with the prediction results of GRAPES_GFS, each numerical evaluation index of the model is improved in the 2 m temperature prediction with an aging of 12-36 hours. The Pearson correlation coefficient increases from about 0.5 to about 0.88, the root mean square error decreases from 1.74-2.06 ℃ to 0.9-1.1 ℃, and the average absolute error decreases from 1.36-1.64℃ to 0.7-0.84℃; At the same time, compared with the mainstream revised models, the model also achieved better correction results.
In order to reduce the high deviation between the predicted value and the observed value of the traditional numerical prediction model (GRAPES_GFS) of 2m temperature and improve the prediction accuracy. Combining the grid data of GRAPES_GFS and the corresponding observation data, this paper proposes a 2m temperature forecast correction model based on Convolution Long Short Time Memory Network (ConvLSTM) and attention mechanism. The model mainly includes the following steps: First, the local shallow features of the input data are extracted by the local feature extraction module. Secondly, input the extracted feature map into the feature attention module, assign different weights to different channel dimensions and different spatial dimensions of the data, restrain the weight of meteorological elements with low correlation with 2m temperature, and achieve local enhancement of high-temperature areas in 2m temperature data. Finally, ConvLSTM network is used to capture the time dimension characteristics of data and output the forecast correction results. The experimental results show that, compared with the prediction results of GRAPES_GFS, each numerical evaluation index of the model is improved in the 2 m temperature prediction with an aging of 12-36 hours. The Pearson correlation coefficient increases from about 0.5 to about 0.88, the root mean square error decreases from 1.74-2.06 ℃ to 0.9-1.1 ℃, and the average absolute error decreases from 1.36-1.64℃ to 0.7-0.84℃; At the same time, compared with the mainstream revised models, the model also achieved better correction results.
Abstract:
To address the issue that convolutional neural networks can hardly balance the local details and global semantic consistency of results in the process of image inpainting, a multi-scale semantic learning model for face image inpainting based on generative adversarial networks is proposed. Firstly, the face image is decomposed into components with different perceptual fields and feature resolutions using gated convolution, and multi-scale features are extracted using convolution kernels of different sizes to enhance the detail of the restoration results by extracting appropriate local features; Secondly, the extracted multi-scale features are fed into the semantic learning module to learn the semantic relationships between features from both the channel and spatial perspectives, thus enhancing the global consistency of the restoration results; finally, skip connections are introduced to complement the features on the encoding side to the decoding side to reduce the loss of detail information caused by sampling and improve the texture details of the restoration results. Experiments on the CelebA-HQ face dataset show that the proposed model has significant improvements in three performance metrics: peak signal to noise ratio, structure similarity, \ell_1, and the inpainting results are visually more reasonable in terms of local details and global semantics.
To address the issue that convolutional neural networks can hardly balance the local details and global semantic consistency of results in the process of image inpainting, a multi-scale semantic learning model for face image inpainting based on generative adversarial networks is proposed. Firstly, the face image is decomposed into components with different perceptual fields and feature resolutions using gated convolution, and multi-scale features are extracted using convolution kernels of different sizes to enhance the detail of the restoration results by extracting appropriate local features; Secondly, the extracted multi-scale features are fed into the semantic learning module to learn the semantic relationships between features from both the channel and spatial perspectives, thus enhancing the global consistency of the restoration results; finally, skip connections are introduced to complement the features on the encoding side to the decoding side to reduce the loss of detail information caused by sampling and improve the texture details of the restoration results. Experiments on the CelebA-HQ face dataset show that the proposed model has significant improvements in three performance metrics: peak signal to noise ratio, structure similarity, \ell_1, and the inpainting results are visually more reasonable in terms of local details and global semantics.
Abstract:
When using global navigation satellite system reflectometry (GNSS-R) signals for tide level inversion, multipath frequencies need to be estimated. The conventional inversion method only estimates the principal frequency, so there are problems such as low data utilization and insufficient temporal resolution of the inversion results. To solve this problem, this paper uses Savitzky-Golay (SG) smoothing filtering to optimize GNSS-R tide level inversion. Firstly, the Lomb-Scargle periodigraph LSP method is used to extract the first four frequencies f1~f4 of signal power, and their corresponding tide level values are inverted. Then, the SG smoothing filter method is used to extract the best inversion results. Finally, 30 days of data from BRST and MAGG stations in France were used to verify the effectiveness of the algorithm. Compared with the LSP method and the window LSP (WINLSP) method, the results show that compared with the LSP method, the average daily inversion values of BRST station and MAGG station after filtering reach 19.30 and 15.23 / day, respectively, which are 34.3% and 19.6% higher. The maximum time interval of the inversion value was 6.63h and 7.07h, respectively, a decrease of 43.2% and 29.4%. RMSE values of 0.3211m and 0.2209m, respectively, are comparable to the LSP method. Compared with the WINLSP method, the average daily inversion values of BRST station and MAGG station after filtering were 60.03/day and 71.17/day, respectively, an increase of 24.2% and 45.9%. The maximum time interval of the inversion value reached 3.08h and 4.50h, a decrease of 25.4% and 28.6%. RMSE values were 0.5222m and 0.3147m, both reduced by 7cm. In general, this method can improve the number of inversion results, improve the utilization rate of data and the temporal resolution of tide level inversion under the premise of ensuring accuracy.
When using global navigation satellite system reflectometry (GNSS-R) signals for tide level inversion, multipath frequencies need to be estimated. The conventional inversion method only estimates the principal frequency, so there are problems such as low data utilization and insufficient temporal resolution of the inversion results. To solve this problem, this paper uses Savitzky-Golay (SG) smoothing filtering to optimize GNSS-R tide level inversion. Firstly, the Lomb-Scargle periodigraph LSP method is used to extract the first four frequencies f1~f4 of signal power, and their corresponding tide level values are inverted. Then, the SG smoothing filter method is used to extract the best inversion results. Finally, 30 days of data from BRST and MAGG stations in France were used to verify the effectiveness of the algorithm. Compared with the LSP method and the window LSP (WINLSP) method, the results show that compared with the LSP method, the average daily inversion values of BRST station and MAGG station after filtering reach 19.30 and 15.23 / day, respectively, which are 34.3% and 19.6% higher. The maximum time interval of the inversion value was 6.63h and 7.07h, respectively, a decrease of 43.2% and 29.4%. RMSE values of 0.3211m and 0.2209m, respectively, are comparable to the LSP method. Compared with the WINLSP method, the average daily inversion values of BRST station and MAGG station after filtering were 60.03/day and 71.17/day, respectively, an increase of 24.2% and 45.9%. The maximum time interval of the inversion value reached 3.08h and 4.50h, a decrease of 25.4% and 28.6%. RMSE values were 0.5222m and 0.3147m, both reduced by 7cm. In general, this method can improve the number of inversion results, improve the utilization rate of data and the temporal resolution of tide level inversion under the premise of ensuring accuracy.
Abstract:
Generative Adversarial Network (GAN) can generate generated images that are close to real images. As a new image generation model in deep learning, GAN plays an important role in image style transfer. Aiming at the problems of low quality of generated images and difficult training of models in the current generation confrontation network models, a new style transfer method is proposed. In this paper, the BicycleGAN model is effectively improved to realize the style transfer of images. Firstly, in order to solve the degradation of GAN in training, the residual module is introduced into the generator of GAN, and the self-attention mechanism is introduced to obtain more image features and improve the generation quality of the generator. In order to solve the gradient explosion phenomenon in the training process of GAN, spectral normalization is added behind each convolution layer of the discriminator. In order to solve the problem of unstable training and low image quality, perceptual loss is introduced. The experimental results on facades and Aerial Photo&Map data sets show that the PSNR and SSIM values of the generated images of this method are higher than those of similar comparison methods.
Generative Adversarial Network (GAN) can generate generated images that are close to real images. As a new image generation model in deep learning, GAN plays an important role in image style transfer. Aiming at the problems of low quality of generated images and difficult training of models in the current generation confrontation network models, a new style transfer method is proposed. In this paper, the BicycleGAN model is effectively improved to realize the style transfer of images. Firstly, in order to solve the degradation of GAN in training, the residual module is introduced into the generator of GAN, and the self-attention mechanism is introduced to obtain more image features and improve the generation quality of the generator. In order to solve the gradient explosion phenomenon in the training process of GAN, spectral normalization is added behind each convolution layer of the discriminator. In order to solve the problem of unstable training and low image quality, perceptual loss is introduced. The experimental results on facades and Aerial Photo&Map data sets show that the PSNR and SSIM values of the generated images of this method are higher than those of similar comparison methods.
Abstract:
Most automatic weather station use optical devices to sample visibility. There are two problems with this. One problem is inaccurate sample date caused by weather factors such as rain, snow or dust in which the lens of those optimal devices might be soiled, and thus the collected data is not accurate. Another problem is data missing due to aging equipment, equipment maintenance, etc. To get rid of those wrong sample and provide complete data for meteorological prediction, this paper proposes a SVM-BP neural network based data quality control method for weather visibility according to those historical data from the Anhui Meteorological Bureau. Firstly, the grey correlation analysis method is used to screen out other meteorological elements closely related to visibility. Secondly, the optimal weight combination is used to combine the support vector machine and the BP neural network. The combined estimation of the visibility of different terrains is then carried out and compared with a single estimation method. The result shows that the combined estimation method has a smaller mean error, higher overall accuracy, and provide an effective theoretical basis for short-term weather forecasting, live analysis, scientific research and public service.
Most automatic weather station use optical devices to sample visibility. There are two problems with this. One problem is inaccurate sample date caused by weather factors such as rain, snow or dust in which the lens of those optimal devices might be soiled, and thus the collected data is not accurate. Another problem is data missing due to aging equipment, equipment maintenance, etc. To get rid of those wrong sample and provide complete data for meteorological prediction, this paper proposes a SVM-BP neural network based data quality control method for weather visibility according to those historical data from the Anhui Meteorological Bureau. Firstly, the grey correlation analysis method is used to screen out other meteorological elements closely related to visibility. Secondly, the optimal weight combination is used to combine the support vector machine and the BP neural network. The combined estimation of the visibility of different terrains is then carried out and compared with a single estimation method. The result shows that the combined estimation method has a smaller mean error, higher overall accuracy, and provide an effective theoretical basis for short-term weather forecasting, live analysis, scientific research and public service.
Abstract:
The rapid development of network technology, human-computer interaction and artificial intelligence has given birth to the metaverse and further promoted the digital transformation of all aspects of people's material life. 2021 is the first year of the metaverse, as a new concept, the metaverse has attracted extensive attention from the industry, academia, media and the public. This paper attempts to deeply analyze the metaverse from the perspective of technology and application. First, the concept and connotation of the metaverse are discussed from the definition, origin and development, characteristics and key technologies (network and computing technology, Internet of Things technology, human-computer interaction technology, electronic game technology, block chain technology, digital twin technology and other technologies); Then, the enterprises and application examples of the current layout of the metaverse are discussed; Finally, the existing problems and opportunities in the development of the metaverse are analyzed, and the future research and application are prospected. Through the induction and analysis of the current development status and research trend of the metaverse and the scientific evaluation of the landing application of the metaverse, we can provide a useful reference for the researchers of the metaverse.
The rapid development of network technology, human-computer interaction and artificial intelligence has given birth to the metaverse and further promoted the digital transformation of all aspects of people's material life. 2021 is the first year of the metaverse, as a new concept, the metaverse has attracted extensive attention from the industry, academia, media and the public. This paper attempts to deeply analyze the metaverse from the perspective of technology and application. First, the concept and connotation of the metaverse are discussed from the definition, origin and development, characteristics and key technologies (network and computing technology, Internet of Things technology, human-computer interaction technology, electronic game technology, block chain technology, digital twin technology and other technologies); Then, the enterprises and application examples of the current layout of the metaverse are discussed; Finally, the existing problems and opportunities in the development of the metaverse are analyzed, and the future research and application are prospected. Through the induction and analysis of the current development status and research trend of the metaverse and the scientific evaluation of the landing application of the metaverse, we can provide a useful reference for the researchers of the metaverse.
Abstract:
To improve the access capacity of distributed photovoltaic (DPV) in the low-voltage distribution station area and promote photovoltaic consumption, a distributionally robust optimization method of DPV access in the low-voltage distribution station area considering the source-load timing characteristics is proposed. First, a source-load joint timing scenario generation method based on optimized clustering is presented to handle the uncertainty of distributed photovoltaic output and load demand in the low-voltage distribution station area; Next, A distributionally robust optimization model of distributed photovoltaic access in the low-voltage distribution station area was constructed by taking into account the voltage constraints, line capacity constraints, reactive power compensation constraints of inverter and photovoltaic consumption constraints, etc. It maximizes the access capacity of DPV while ensuring that the expected value of abandoned light rate under the worst probability distribution of each typical scenario meets the requirements. Then, a mathematical model of distributed energy storage access in low-voltage distribution stations area is established to study the influence of energy storage access and its charging-discharging mechanism on distributed photovoltaic access. Finally, the effectiveness of the model in this paper is verified by taking the actual low-voltage distribution station area as an example.
To improve the access capacity of distributed photovoltaic (DPV) in the low-voltage distribution station area and promote photovoltaic consumption, a distributionally robust optimization method of DPV access in the low-voltage distribution station area considering the source-load timing characteristics is proposed. First, a source-load joint timing scenario generation method based on optimized clustering is presented to handle the uncertainty of distributed photovoltaic output and load demand in the low-voltage distribution station area; Next, A distributionally robust optimization model of distributed photovoltaic access in the low-voltage distribution station area was constructed by taking into account the voltage constraints, line capacity constraints, reactive power compensation constraints of inverter and photovoltaic consumption constraints, etc. It maximizes the access capacity of DPV while ensuring that the expected value of abandoned light rate under the worst probability distribution of each typical scenario meets the requirements. Then, a mathematical model of distributed energy storage access in low-voltage distribution stations area is established to study the influence of energy storage access and its charging-discharging mechanism on distributed photovoltaic access. Finally, the effectiveness of the model in this paper is verified by taking the actual low-voltage distribution station area as an example.
Abstract:
Speaker recognition has become a second ID card for people. In order to meet the application needs of speaker recognition for short-time speech in open scenarios, the paper will optimize the model of speaker recognition to improve the accuracy and robustness of the model. First, in order to realize the selection of important frequency features in the input acoustic features, Reweighted-based Feature Enhancement Layer (RFEL) and Reweighted-based Feature Enhancement Network (RFEN) are proposed to enhance the feature expression, which serves to enhance feature representation. Second, the loss function of Mis-Classified Vector Guided Softmax Loss(MV) in face recognition is introduced to the speaker recognition field for the first time to improve the mining ability of hard samples. Third, we propose a combined loss function based on MV and Angular Prototypical Loss (AP) based on a few-shot framework, which solves the mismatch between the classification loss function and the actual evaluation requirements of speaker recognition. Besises, the strong dependence of the metric function on the sampling strategy is also solved. Finally, The experimental results show that the performance metric EER of the model proposed in this paper is reduced by 12.45% compared to the baseline, and the minDCF is relatively reduced by 14.09%, achieving excellent results in the existing speaker recognition field.
Speaker recognition has become a second ID card for people. In order to meet the application needs of speaker recognition for short-time speech in open scenarios, the paper will optimize the model of speaker recognition to improve the accuracy and robustness of the model. First, in order to realize the selection of important frequency features in the input acoustic features, Reweighted-based Feature Enhancement Layer (RFEL) and Reweighted-based Feature Enhancement Network (RFEN) are proposed to enhance the feature expression, which serves to enhance feature representation. Second, the loss function of Mis-Classified Vector Guided Softmax Loss(MV) in face recognition is introduced to the speaker recognition field for the first time to improve the mining ability of hard samples. Third, we propose a combined loss function based on MV and Angular Prototypical Loss (AP) based on a few-shot framework, which solves the mismatch between the classification loss function and the actual evaluation requirements of speaker recognition. Besises, the strong dependence of the metric function on the sampling strategy is also solved. Finally, The experimental results show that the performance metric EER of the model proposed in this paper is reduced by 12.45% compared to the baseline, and the minDCF is relatively reduced by 14.09%, achieving excellent results in the existing speaker recognition field.
Abstract:
Abstract: The accuracy is an important indicator of the quality of UAV oblique photography three-dimensional(3D) model. Conventional modeling steps need to deploy ground control points (GCPs), which is inefficient and costly. Control-free modeling does not require GCPs, which is efficiency but low accuracy. Thus, this paper proposed a 3D modeling method of oblique photography with post-processed image control. The study area is in the east area of Nanjing University of Information Science and Technology. The results show that: 1) The accuracy of the 3D models established by the two methods is basically the same, and the plane error and the elevation error are both less than 0.05m, which meets the modeling requirements of high efficiency, high precision and low cost; 2) the photograph control post-processed modeling avoids the process of making and maintaining the photograph control point. photograph control points are selected on the existing high-precision 3D model and the coordinates are extracted, which improves the scientificity of the photograph control point selection and reduces the workload of setting GCPs; 3) This method has good universality in the regular updating of high-precision 3D real scene models in cities, towns, roads and other areas. Keywords: oblique photography; photograph control point; 3D reality model; photograph control post-processed; precision
Abstract: The accuracy is an important indicator of the quality of UAV oblique photography three-dimensional(3D) model. Conventional modeling steps need to deploy ground control points (GCPs), which is inefficient and costly. Control-free modeling does not require GCPs, which is efficiency but low accuracy. Thus, this paper proposed a 3D modeling method of oblique photography with post-processed image control. The study area is in the east area of Nanjing University of Information Science and Technology. The results show that: 1) The accuracy of the 3D models established by the two methods is basically the same, and the plane error and the elevation error are both less than 0.05m, which meets the modeling requirements of high efficiency, high precision and low cost; 2) the photograph control post-processed modeling avoids the process of making and maintaining the photograph control point. photograph control points are selected on the existing high-precision 3D model and the coordinates are extracted, which improves the scientificity of the photograph control point selection and reduces the workload of setting GCPs; 3) This method has good universality in the regular updating of high-precision 3D real scene models in cities, towns, roads and other areas. Keywords: oblique photography; photograph control point; 3D reality model; photograph control post-processed; precision
Abstract:
In order to solve the problem of poor modeling effect caused by many shooting blind areas and large differences in image acquisition resolution in complex buildings, a three-dimensional surround photography and fine modeling method of UAV is proposed. A single lens small multi rotor UAV is selected to design a three-dimensional surround image acquisition route according to the appearance of complex buildings, Then import the acquired image into context capture software for analysis and processing, and complete the model establishment. Finally, aiming at the lack of texture, fuzzy drawing and other information in the three-dimensional model, DP modeler and 3ds Max are used to modify the model. Taking two adjacent buildings in a park as the research object, the new method and the traditional UAV tilt photography method are used to carry out comparative experiments from three aspects: modeling efficiency, model accuracy and texture details. The results show that the new method has the advantages of high modeling efficiency, high precision and rich texture, improves the quality and efficiency of complex building modeling, linkage modification solves the problem of lack of information in the process of image acquisition, and significantly improves the quality of model details.
In order to solve the problem of poor modeling effect caused by many shooting blind areas and large differences in image acquisition resolution in complex buildings, a three-dimensional surround photography and fine modeling method of UAV is proposed. A single lens small multi rotor UAV is selected to design a three-dimensional surround image acquisition route according to the appearance of complex buildings, Then import the acquired image into context capture software for analysis and processing, and complete the model establishment. Finally, aiming at the lack of texture, fuzzy drawing and other information in the three-dimensional model, DP modeler and 3ds Max are used to modify the model. Taking two adjacent buildings in a park as the research object, the new method and the traditional UAV tilt photography method are used to carry out comparative experiments from three aspects: modeling efficiency, model accuracy and texture details. The results show that the new method has the advantages of high modeling efficiency, high precision and rich texture, improves the quality and efficiency of complex building modeling, linkage modification solves the problem of lack of information in the process of image acquisition, and significantly improves the quality of model details.
Abstract:
This paper studies the optimization and coordination of the low-carbon supply chain considering triple bottom line under carbon tax policy. First, we study the decisions of the centralized supply chain considering the triple bottom line. Secondly, we consider four different decentralized supply chain models under the wholesale price contract, i.e., Model I, Model II, Model III, and Model IV. Then we study four different decentralized low-carbon supply chain models under two-part tariff contracts, i.e., Model I-LTT, Model II-LTT, Model III-LTT, and Model IV-LTT. The results show that the two-part tariff contracts can coordinate the low-carbon supply chain perfectly in Model I-LTT and Model III-LTT. At the same time, the low-carbon supply chain in Model II-LTT and Model IV-LTT can realize Pareto improvement. Finally, we conduct the numerical analysis to study the impacts of carbon tax and carbon emission reduction (CER) investment coefficient on the supply chain profits before and after coordination. This paper provides the reliable theoretical basis for low-carbon enterprises to choose CER strategies and contracts.
This paper studies the optimization and coordination of the low-carbon supply chain considering triple bottom line under carbon tax policy. First, we study the decisions of the centralized supply chain considering the triple bottom line. Secondly, we consider four different decentralized supply chain models under the wholesale price contract, i.e., Model I, Model II, Model III, and Model IV. Then we study four different decentralized low-carbon supply chain models under two-part tariff contracts, i.e., Model I-LTT, Model II-LTT, Model III-LTT, and Model IV-LTT. The results show that the two-part tariff contracts can coordinate the low-carbon supply chain perfectly in Model I-LTT and Model III-LTT. At the same time, the low-carbon supply chain in Model II-LTT and Model IV-LTT can realize Pareto improvement. Finally, we conduct the numerical analysis to study the impacts of carbon tax and carbon emission reduction (CER) investment coefficient on the supply chain profits before and after coordination. This paper provides the reliable theoretical basis for low-carbon enterprises to choose CER strategies and contracts.
Abstract:
The change of daily precipitation presents significant non-linear characteristics, and it is very difficult to accurately predict it. In recent years, Long-Short Term Memory (LSTM) has obvious advantages in precipitation prediction. However, the deep structure of LSTM causes its shortcomings such as over-fitting and time lag, which affect the prediction accuracy. Note that the Broad Learning System (BLS) directly calculates weights, and its feature of not requiring multiple iterations can help solve the shortcomings of LSTM. However, noise and outliers still have an adverse effect on modeling. Based on this, an improved Weighted Broad Learning System (WBLS) is proposed. Use a weighted penalty factor to constrain each sample, and assign high and low weights to normal and abnormal samples to increase and decrease their influence. This paper combines the advantages of LSTM and WBLS to propose a LSTM-WBLS daily precipitation prediction model. Empirical research is carried out on the actual data of daily precipitation at Badong Station in Hubei Province in the past 20 years, and factors such as air pressure, temperature, humidity, wind speed and sunshine are taken into consideration. The results show that, compared with the existing prediction models, the LSTM-BLS model has the highest prediction accuracy in all evaluation indicators. In particular, the WBLS module has been added to solve the time lag problem of LSTM. Moreover, under different time steps, the prediction accuracy of the new model also performed best, proving its stability. In terms of computational efficiency, LSTM-WBLS has not decreased compared with LSTM.
The change of daily precipitation presents significant non-linear characteristics, and it is very difficult to accurately predict it. In recent years, Long-Short Term Memory (LSTM) has obvious advantages in precipitation prediction. However, the deep structure of LSTM causes its shortcomings such as over-fitting and time lag, which affect the prediction accuracy. Note that the Broad Learning System (BLS) directly calculates weights, and its feature of not requiring multiple iterations can help solve the shortcomings of LSTM. However, noise and outliers still have an adverse effect on modeling. Based on this, an improved Weighted Broad Learning System (WBLS) is proposed. Use a weighted penalty factor to constrain each sample, and assign high and low weights to normal and abnormal samples to increase and decrease their influence. This paper combines the advantages of LSTM and WBLS to propose a LSTM-WBLS daily precipitation prediction model. Empirical research is carried out on the actual data of daily precipitation at Badong Station in Hubei Province in the past 20 years, and factors such as air pressure, temperature, humidity, wind speed and sunshine are taken into consideration. The results show that, compared with the existing prediction models, the LSTM-BLS model has the highest prediction accuracy in all evaluation indicators. In particular, the WBLS module has been added to solve the time lag problem of LSTM. Moreover, under different time steps, the prediction accuracy of the new model also performed best, proving its stability. In terms of computational efficiency, LSTM-WBLS has not decreased compared with LSTM.
Abstract:
For the problem of low efficiency and high false alarm rate of DDoS (Distributed denial of service) network traffic attack detection, this paper proposes a DDoS attack detection method based on Discrete Wavelet Transform DWT and Adaptive Knowledge Distillation AKD. Distillation AKD) self-encoder neural network based DDoS attack detection method. The method uses discrete wavelet transform to extract frequency features, the auto-encoder neural network to encode and classify the features, and the adaptive knowledge distillation to compress the model in order to achieve efficient detection of DDoS attack traffic. The research results show that the method has high detection efficiency for proxy server attacks, database vulnerabilities and TCP flood attacks, UDP flood attacks, and has a low false alarm rate.
For the problem of low efficiency and high false alarm rate of DDoS (Distributed denial of service) network traffic attack detection, this paper proposes a DDoS attack detection method based on Discrete Wavelet Transform DWT and Adaptive Knowledge Distillation AKD. Distillation AKD) self-encoder neural network based DDoS attack detection method. The method uses discrete wavelet transform to extract frequency features, the auto-encoder neural network to encode and classify the features, and the adaptive knowledge distillation to compress the model in order to achieve efficient detection of DDoS attack traffic. The research results show that the method has high detection efficiency for proxy server attacks, database vulnerabilities and TCP flood attacks, UDP flood attacks, and has a low false alarm rate.
Abstract:
This paper uses the GNSS dual-frequency observation data of the BRST port and the Severn Bridge monitoring system in the United Kingdom to perform GPS-IR water level inversion in static and high dynamic environments, respectively. To explore the feasibility and accuracy of water level inversion by traditional GNSS monitoring system in static and high dynamic scenarios. The results show that the inversion accuracy of the L1 band is higher than that of the L2 band; in the static scene, the correlation coefficient between the GPS-IR water level inversion results and the tide gauge data is above 0.98; in the high dynamic scene, the retrieval accuracy of F001 station is slightly lower, using the Empirical Mode Decomposition (EMD) method to improve the algorithm, improve the accuracy of GPS-IR water level retrieval results in complex bridge environment, compared with the classical method, Root Mean Square Error is reduced by about 50%. Improve the applicability of GPS-IR technology in different water environments, the results show that GPS-IR technology has a good application prospect in water level monitoring.
This paper uses the GNSS dual-frequency observation data of the BRST port and the Severn Bridge monitoring system in the United Kingdom to perform GPS-IR water level inversion in static and high dynamic environments, respectively. To explore the feasibility and accuracy of water level inversion by traditional GNSS monitoring system in static and high dynamic scenarios. The results show that the inversion accuracy of the L1 band is higher than that of the L2 band; in the static scene, the correlation coefficient between the GPS-IR water level inversion results and the tide gauge data is above 0.98; in the high dynamic scene, the retrieval accuracy of F001 station is slightly lower, using the Empirical Mode Decomposition (EMD) method to improve the algorithm, improve the accuracy of GPS-IR water level retrieval results in complex bridge environment, compared with the classical method, Root Mean Square Error is reduced by about 50%. Improve the applicability of GPS-IR technology in different water environments, the results show that GPS-IR technology has a good application prospect in water level monitoring.
Abstract:
There are quite a few comment sentences without emotional words in aspect-level emotional texts, and the study of their emotions is called aspect-level implicit emotional analysis. The existing models have the problems that the context information related to aspect words may be lost in the pre-training process, and the deep features in the context cannot be accurately extracted. Aiming at the first problem, this paper constructs a aspect-aware BERT pre-training model, and introduces aspect words into the input embedding structure of basic BERT to generate word vectors related to aspect words. Aiming at the second problem, this paper constructs a context-aware attention mechanism. For the deep hidden vectors obtained from the coding layer, the semantic and syntactic information is introduced into the attention weight calculation process, so that the attention mechanism can more accurately assign the weight to the context related to aspect words. The results of comparative experiments show that the effect of this model is better than that of the baseline model.
There are quite a few comment sentences without emotional words in aspect-level emotional texts, and the study of their emotions is called aspect-level implicit emotional analysis. The existing models have the problems that the context information related to aspect words may be lost in the pre-training process, and the deep features in the context cannot be accurately extracted. Aiming at the first problem, this paper constructs a aspect-aware BERT pre-training model, and introduces aspect words into the input embedding structure of basic BERT to generate word vectors related to aspect words. Aiming at the second problem, this paper constructs a context-aware attention mechanism. For the deep hidden vectors obtained from the coding layer, the semantic and syntactic information is introduced into the attention weight calculation process, so that the attention mechanism can more accurately assign the weight to the context related to aspect words. The results of comparative experiments show that the effect of this model is better than that of the baseline model.
Abstract:
The prediction of reservoir water level provides important decision support for its operation, flood control and water resources operation and management.Accurate and reliable prediction plays an important role in the optimal management of water resources. Hydrodynamic prediction method requires high input boundary conditions and large prediction error,which is difficult to meet the requirements of real-time water level regulation.Aiming at the nonlinearity, instability and complex temporal and spatial characteristics of reservoir water level data,a hybrid reservoir water level prediction model integrating adaptive Variational Modal Decomposition (VMD),Convolutional Neural Networks(CNN) and Gated Recurrent Unit (GRU) is proposed. Among them,VMD eliminates noise by decomposing the water level sequence,CNN is used to effectively extract the local features of water level data,and GRU is used to extract the deep time features of water level data.Taking the daily water level prediction of Shenwo reservoir as an example,compared with several related models,the analysis results show that the new model performs best in the selected evaluation indexes. Among them, the Root Mean Square Error is 0.1500, the Mean Absolute Error is 0.1063,the Mean Absolute Percentage Error is 0.1145,and the Nash–Sutcliffe efficiency coefficient is as high as 0.9953.In particular,the operation efficiency of GRU selected in this paper is significantly improved compared with Long and Short-term Memory network(LSTM). Therefore,the new model has high accuracy and high operation efficiency, and is more suitable for the real-time operation of reservoir water level.
The prediction of reservoir water level provides important decision support for its operation, flood control and water resources operation and management.Accurate and reliable prediction plays an important role in the optimal management of water resources. Hydrodynamic prediction method requires high input boundary conditions and large prediction error,which is difficult to meet the requirements of real-time water level regulation.Aiming at the nonlinearity, instability and complex temporal and spatial characteristics of reservoir water level data,a hybrid reservoir water level prediction model integrating adaptive Variational Modal Decomposition (VMD),Convolutional Neural Networks(CNN) and Gated Recurrent Unit (GRU) is proposed. Among them,VMD eliminates noise by decomposing the water level sequence,CNN is used to effectively extract the local features of water level data,and GRU is used to extract the deep time features of water level data.Taking the daily water level prediction of Shenwo reservoir as an example,compared with several related models,the analysis results show that the new model performs best in the selected evaluation indexes. Among them, the Root Mean Square Error is 0.1500, the Mean Absolute Error is 0.1063,the Mean Absolute Percentage Error is 0.1145,and the Nash–Sutcliffe efficiency coefficient is as high as 0.9953.In particular,the operation efficiency of GRU selected in this paper is significantly improved compared with Long and Short-term Memory network(LSTM). Therefore,the new model has high accuracy and high operation efficiency, and is more suitable for the real-time operation of reservoir water level.
Abstract:
With the wide application of communication technology in the smart grid, the threat of information attack to the safe operation of the grid also follows. Attacks based on state estimation using phasor measurement techniques in power systems are difficult to detect successfully. Aiming at this problem, this paper proposes an intelligent attack detection method based on phasor measurement data state estimation for power systems. In this method, the self-encoder is used to extract the features of the power grid measurement data, and the feature dimension is gradually reduced by extracting the features multiple times. The final extracted information is subjected to supervised learning through the softmax layer, resulting in an attack detection algorithm based on stacked autoencoders. Aiming at the overfitting problem of autoencoders, an attack detection method based on noise reduction autoencoders is further proposed. The proposed method is simulated and verified by IEEE-118 node test system. The results show that the proposed attack detection method has higher computational accuracy and efficiency than other methods.
With the wide application of communication technology in the smart grid, the threat of information attack to the safe operation of the grid also follows. Attacks based on state estimation using phasor measurement techniques in power systems are difficult to detect successfully. Aiming at this problem, this paper proposes an intelligent attack detection method based on phasor measurement data state estimation for power systems. In this method, the self-encoder is used to extract the features of the power grid measurement data, and the feature dimension is gradually reduced by extracting the features multiple times. The final extracted information is subjected to supervised learning through the softmax layer, resulting in an attack detection algorithm based on stacked autoencoders. Aiming at the overfitting problem of autoencoders, an attack detection method based on noise reduction autoencoders is further proposed. The proposed method is simulated and verified by IEEE-118 node test system. The results show that the proposed attack detection method has higher computational accuracy and efficiency than other methods.
Abstract:
It has always been a difficult subject for marine radar to detect small targets on sea surface. To overcome the low detection probability of traditional detectors, a feature detector based on relative sample entropy (denoted as FD-RSE) is proposed in this paper. Firstly, the whitened spectrum is defined to suppress the main clutter region, thus enlarging the irregularity of the sea clutter sequence. Then, by introducing sample entropy describe the complexity of sea clutter sequence, relative sample entropy is extracted from whitened spectrum to serve as feature. Therefore, the difference between the geometric characteristic of sea clutter and that of target echo can be thoroughly exploited in the Doppler spectrum. Finally, the superiority of the proposed FD-RSE detector over traditional detectors in improving detection performance can be verified by the IPIX measured dataset.
It has always been a difficult subject for marine radar to detect small targets on sea surface. To overcome the low detection probability of traditional detectors, a feature detector based on relative sample entropy (denoted as FD-RSE) is proposed in this paper. Firstly, the whitened spectrum is defined to suppress the main clutter region, thus enlarging the irregularity of the sea clutter sequence. Then, by introducing sample entropy describe the complexity of sea clutter sequence, relative sample entropy is extracted from whitened spectrum to serve as feature. Therefore, the difference between the geometric characteristic of sea clutter and that of target echo can be thoroughly exploited in the Doppler spectrum. Finally, the superiority of the proposed FD-RSE detector over traditional detectors in improving detection performance can be verified by the IPIX measured dataset.
Abstract:
Due to the diversity of the distribution and shape of rain streaks, existing rain removal algorithms produce problems such as blurred image background and poor generalization performance while removing rain. A multi-scale feature fusion image rain removal method based on an attention mechanism is proposed to address these problems. The feature extraction stage consists of multiple residual groups containing two multi-scale attention residual blocks. The multi-scale attention residual blocks use the multi-scale feature extraction module to extract and aggregate feature information at different scales and further improve the feature extraction capability of the network through coordinate attention. Local feature fusion is performed within groups, and the global feature fusion attention module is used between groups to better fuse features at different levels and to focus the network on rain streaks regions through pixel attention. The quantitative metrics of the proposed method are significantly improved compared with other existing image rain removal algorithms on both simulated and real rain image datasets, and the visual effects of the rain removal images are better and have good generalization.
Due to the diversity of the distribution and shape of rain streaks, existing rain removal algorithms produce problems such as blurred image background and poor generalization performance while removing rain. A multi-scale feature fusion image rain removal method based on an attention mechanism is proposed to address these problems. The feature extraction stage consists of multiple residual groups containing two multi-scale attention residual blocks. The multi-scale attention residual blocks use the multi-scale feature extraction module to extract and aggregate feature information at different scales and further improve the feature extraction capability of the network through coordinate attention. Local feature fusion is performed within groups, and the global feature fusion attention module is used between groups to better fuse features at different levels and to focus the network on rain streaks regions through pixel attention. The quantitative metrics of the proposed method are significantly improved compared with other existing image rain removal algorithms on both simulated and real rain image datasets, and the visual effects of the rain removal images are better and have good generalization.
Abstract:
The development process and characteristic analysis of precipitation cloud system is an important problem in the field of cloud precipitation physics. In this paper, we select the 700 hPa cloud water content (CWC) and the 1h value of airflow velocity (OMG) in the vertical direction of the atmosphere, measure the chaos degree of CWC distribution with the information entropy as a tool, and OMG to judge the development of cloud, a combined prediction model is also proposed based on hybrid multi-scale decomposition, Holt model, autoregressive integrated moving average model (ARIMA) and Lagrange Multiplier. Results show that: 1) The CWC entropy has non-linear and non-stationary characteristics; 2) At the different development stages of the cloud, The means of the CWC entropy sequence of northern clouds are all smaller than those of southern clouds, The variance is generally greater than that of the southern cloud; 3) The OMG mean and the extreme point of the wavelet low-frequency reconstruction of the CWC entropy correspond well in time, Close extreme value points account for 50% of the southern cloud, In 83.3% of the northern cloud, It shows that CWC entropy can reflect the development of cloud system to a certain extent; 4) CWC entropy sequences often have multiple timescale features, Therefore, the accuracy of the Holt-ARIMA-Lagrange Multiplier model after multi-scale decomposition is improved by more than 3%, compared with the single prediction method and the prediction model of single-layer decomposition.
The development process and characteristic analysis of precipitation cloud system is an important problem in the field of cloud precipitation physics. In this paper, we select the 700 hPa cloud water content (CWC) and the 1h value of airflow velocity (OMG) in the vertical direction of the atmosphere, measure the chaos degree of CWC distribution with the information entropy as a tool, and OMG to judge the development of cloud, a combined prediction model is also proposed based on hybrid multi-scale decomposition, Holt model, autoregressive integrated moving average model (ARIMA) and Lagrange Multiplier. Results show that: 1) The CWC entropy has non-linear and non-stationary characteristics; 2) At the different development stages of the cloud, The means of the CWC entropy sequence of northern clouds are all smaller than those of southern clouds, The variance is generally greater than that of the southern cloud; 3) The OMG mean and the extreme point of the wavelet low-frequency reconstruction of the CWC entropy correspond well in time, Close extreme value points account for 50% of the southern cloud, In 83.3% of the northern cloud, It shows that CWC entropy can reflect the development of cloud system to a certain extent; 4) CWC entropy sequences often have multiple timescale features, Therefore, the accuracy of the Holt-ARIMA-Lagrange Multiplier model after multi-scale decomposition is improved by more than 3%, compared with the single prediction method and the prediction model of single-layer decomposition.
Abstract:
To examine the effects of gradual increase of atmospheric CO2 concentration on soil respiration of winter wheat (Triticum aestivum) field, a gradually increased CO2 concentration experiment was conducted with automatic control system of CO2 in open top chambers (OTCs) during 2017?2019 growing seasons. In this study, a gradual increase of atmospheric CO2 concentration (C80 and C120, an increase of 40 μmol·mol?1 year by year from 2016) was set up based on the ambient atmospheric CO2 concentration (CK). The soil respiration rate (Rs) was measured by the static chamber?gas chromatograph method. The results showed that gradually increased CO2 did not alter the seasonal patterns of soil respiration, but had significant effect on Rs during winter wheat bloom?growth period. In 2018?2019 growing season, compared to CK, C120 treatment significantly increased Rs by 50.2% (P=0.008) at the heading?flowering stage, and significantly increased cumulative amount of CO2 emissions (CAC) by 25.9% (P=0.044) during the wheat growing season, but in 2017?2018 growing season, compared to CK, C80 treatment had no significant effect on Rs. Soil respiration showed a positive exponential and negative linear correlation with the soil temperature and soil moisture, respectively. Compared to CK, gradually increased CO2 concentration reduced the temperature sensitivity coefficient of soil respiration (Q10 values). In summary, a gradual increase of atmospheric CO2 concentration to 120 μmol·mol?1 increased CAC during the growing season of winter wheat.
To examine the effects of gradual increase of atmospheric CO2 concentration on soil respiration of winter wheat (Triticum aestivum) field, a gradually increased CO2 concentration experiment was conducted with automatic control system of CO2 in open top chambers (OTCs) during 2017?2019 growing seasons. In this study, a gradual increase of atmospheric CO2 concentration (C80 and C120, an increase of 40 μmol·mol?1 year by year from 2016) was set up based on the ambient atmospheric CO2 concentration (CK). The soil respiration rate (Rs) was measured by the static chamber?gas chromatograph method. The results showed that gradually increased CO2 did not alter the seasonal patterns of soil respiration, but had significant effect on Rs during winter wheat bloom?growth period. In 2018?2019 growing season, compared to CK, C120 treatment significantly increased Rs by 50.2% (P=0.008) at the heading?flowering stage, and significantly increased cumulative amount of CO2 emissions (CAC) by 25.9% (P=0.044) during the wheat growing season, but in 2017?2018 growing season, compared to CK, C80 treatment had no significant effect on Rs. Soil respiration showed a positive exponential and negative linear correlation with the soil temperature and soil moisture, respectively. Compared to CK, gradually increased CO2 concentration reduced the temperature sensitivity coefficient of soil respiration (Q10 values). In summary, a gradual increase of atmospheric CO2 concentration to 120 μmol·mol?1 increased CAC during the growing season of winter wheat.
Abstract:
The ground-level ozone (O3) concentration is continuously increasing in middle areas of China during recent years. Volatile organic compounds (VOCs) are important precursors in photochemical production of ozone, which is of great significance to carrying out research for controlling O3 prevention and control. In order to understand characteristics and role in ozone formation of ambient VOCs in middle areas, we conducted online VOCs observation in Changsha City from May to October in 2021, monitoring 117 species in total. The mean value of mixing ratios for measured VOCs was (23.09±9.97)×10-9. Monthly average concentration of VOCs in Changsha showed a “U” shape, with the lowest in July and the highest in October, while the diurnal variation of VOCs concentration showed a bimodal pattern, indicating the influence of human activities. OVOCs was the most abundant component of VOCs, followed by alkane and halo carbon. However, OVOCs and aromatics were the two largest contributors to ozone formation potential (OFP), with a cumulative contribution of 68.3%. Propionaldehyde, acetaldehyde, m/p-xylene, ethylene and toluene were the most important VOCs species to OFP. The observation-based model(OBM) showed that O3 formation in Changsha was in a transition regime in May, August, September and in a VOCs-limited regime from June to July, while in a NOx-limited regime in October. For the anthropogenic VOCs, ALDP, OLEP and PARP should be given priority in emission control measures.
The ground-level ozone (O3) concentration is continuously increasing in middle areas of China during recent years. Volatile organic compounds (VOCs) are important precursors in photochemical production of ozone, which is of great significance to carrying out research for controlling O3 prevention and control. In order to understand characteristics and role in ozone formation of ambient VOCs in middle areas, we conducted online VOCs observation in Changsha City from May to October in 2021, monitoring 117 species in total. The mean value of mixing ratios for measured VOCs was (23.09±9.97)×10-9. Monthly average concentration of VOCs in Changsha showed a “U” shape, with the lowest in July and the highest in October, while the diurnal variation of VOCs concentration showed a bimodal pattern, indicating the influence of human activities. OVOCs was the most abundant component of VOCs, followed by alkane and halo carbon. However, OVOCs and aromatics were the two largest contributors to ozone formation potential (OFP), with a cumulative contribution of 68.3%. Propionaldehyde, acetaldehyde, m/p-xylene, ethylene and toluene were the most important VOCs species to OFP. The observation-based model(OBM) showed that O3 formation in Changsha was in a transition regime in May, August, September and in a VOCs-limited regime from June to July, while in a NOx-limited regime in October. For the anthropogenic VOCs, ALDP, OLEP and PARP should be given priority in emission control measures.
Abstract:
In recent years, an increasing number of researchers have attempted to link catalytic activity to the acidity and basicity of the oxygen vacancies in catalysts. A deeper understanding of the role played by oxygen vacancy defects in determining catalytic properties and charge carrier dynamics by combining Lewis acid-base active sites can provide valuable insights for the rational development of more efficient catalysts. Oxygen vacancies are the most common defect in transition metal oxides, and catalysts containing oxygen vacancies are closely related to how surface acidity and basicity affect the adsorption, reaction, and desorption processes of reactants, intermediates, and solid surface products. To this end, the location of oxygen vacancies and the acid-base of oxygen vacancies have always played a central role in the history of multiphase catalysis, and the surface properties of multiphase catalysts will largely determine the electronic structure of the surface active sites, thus significantly affecting their corresponding catalytic efficiency, stability, and chemoselectivity. The study of the construction of Lewis acid-base sites based on oxygen vacancy defects in catalysts is of great significance for the performance enhancement of various molecular activation and catalytic conversion reactions. In this paper, we review the research on oxygen vacancy-containing solid catalysts for the construction of Lewis acid-base sites and summarize their relevant applications in environmental catalysis. This paper is a review of the research on the oxygen vacancy-containing solid catalysts for the construction of Lewis acid-base sites and summarizes their relevant applications in the field of photochemical catalysis.
In recent years, an increasing number of researchers have attempted to link catalytic activity to the acidity and basicity of the oxygen vacancies in catalysts. A deeper understanding of the role played by oxygen vacancy defects in determining catalytic properties and charge carrier dynamics by combining Lewis acid-base active sites can provide valuable insights for the rational development of more efficient catalysts. Oxygen vacancies are the most common defect in transition metal oxides, and catalysts containing oxygen vacancies are closely related to how surface acidity and basicity affect the adsorption, reaction, and desorption processes of reactants, intermediates, and solid surface products. To this end, the location of oxygen vacancies and the acid-base of oxygen vacancies have always played a central role in the history of multiphase catalysis, and the surface properties of multiphase catalysts will largely determine the electronic structure of the surface active sites, thus significantly affecting their corresponding catalytic efficiency, stability, and chemoselectivity. The study of the construction of Lewis acid-base sites based on oxygen vacancy defects in catalysts is of great significance for the performance enhancement of various molecular activation and catalytic conversion reactions. In this paper, we review the research on oxygen vacancy-containing solid catalysts for the construction of Lewis acid-base sites and summarize their relevant applications in environmental catalysis. This paper is a review of the research on the oxygen vacancy-containing solid catalysts for the construction of Lewis acid-base sites and summarizes their relevant applications in the field of photochemical catalysis.
Abstract:
While traditional visual SLAM has achieved amazing results, it is difficult to achieve the desired results in challenging environments.Deep learning has promoted the rapid development of computer vision. Combining deep learning with vision-based SLAM is a hot topic. Starting from the classical neural network of deep learning, this paper introduces the combination of deep learning and traditional vision-based SLAM algorithm.The achievements of Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) in depth estimation, pose estimation and closed-loop detection are summarized.The advantages of neural network in semantic information extraction are pointed out, which can help the autonomous mobile robot to be truly independent in the future, and the development of VSLAM in the future is prospected.
While traditional visual SLAM has achieved amazing results, it is difficult to achieve the desired results in challenging environments.Deep learning has promoted the rapid development of computer vision. Combining deep learning with vision-based SLAM is a hot topic. Starting from the classical neural network of deep learning, this paper introduces the combination of deep learning and traditional vision-based SLAM algorithm.The achievements of Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) in depth estimation, pose estimation and closed-loop detection are summarized.The advantages of neural network in semantic information extraction are pointed out, which can help the autonomous mobile robot to be truly independent in the future, and the development of VSLAM in the future is prospected.
Abstract:
In order to suppress the cogging torque of permanent magnet synchronous motor and reduce the vibration and noise of motor, a structure of stator tooth jacking auxiliary slot and rotor outer circle eccentricity is proposed to suppress the cogging torque and improve the output performance of motor. The analytical formula of the distribution function of the stator tooth top before and after slotting and the main air gap is established. The relationship between the stator slotting and the main magnetic field distribution and the cogging torque of the motor is analyzed. At the same time, the influence of the rotor outer circle eccentricity on the cogging torque is analyzed. Taking the built-in 3-phase 8-pole 48 slot V-type permanent magnet synchronous motor as an example, the number of auxiliary slots on the stator tooth top, slot width, slot depth, slot spacing and rotor outer circle eccentricity are optimized and compared by using the finite element method, and the optimal parameter matching is obtained. The results show that the double rectangular symmetrical slotting on the stator tooth top and the eccentricity of the rotor outer circle can effectively improve the magnetic field distribution of the main air gap and restrain the cogging torque of the motor. The harmonic amplitude of the optimized air gap magnetic density 5, 11, 15 and 17 decreases significantly, the peak value of the cogging torque of the motor decreases by 59.6%, the sine of the back EMF waveform increases, the average torque increases, and the output performance of the motor is significantly improved.
In order to suppress the cogging torque of permanent magnet synchronous motor and reduce the vibration and noise of motor, a structure of stator tooth jacking auxiliary slot and rotor outer circle eccentricity is proposed to suppress the cogging torque and improve the output performance of motor. The analytical formula of the distribution function of the stator tooth top before and after slotting and the main air gap is established. The relationship between the stator slotting and the main magnetic field distribution and the cogging torque of the motor is analyzed. At the same time, the influence of the rotor outer circle eccentricity on the cogging torque is analyzed. Taking the built-in 3-phase 8-pole 48 slot V-type permanent magnet synchronous motor as an example, the number of auxiliary slots on the stator tooth top, slot width, slot depth, slot spacing and rotor outer circle eccentricity are optimized and compared by using the finite element method, and the optimal parameter matching is obtained. The results show that the double rectangular symmetrical slotting on the stator tooth top and the eccentricity of the rotor outer circle can effectively improve the magnetic field distribution of the main air gap and restrain the cogging torque of the motor. The harmonic amplitude of the optimized air gap magnetic density 5, 11, 15 and 17 decreases significantly, the peak value of the cogging torque of the motor decreases by 59.6%, the sine of the back EMF waveform increases, the average torque increases, and the output performance of the motor is significantly improved.
Abstract:
In response to the severe haze pollution has become one of the major obstacles to the ecological integration development of the Yangtze River Economic Zone at presenta three-stage DEA model was used to study the haze management efficiency of 27 cities in the Yangtze River Delta from 2014 to 2019. The study shows that (1)Three-stage DEA model can effectively eliminate the effects of environmental factors and random errors, and can truly reflect the level of haze control efficiency in the Yangtze River Delta, and the difference between the integrated efficiency in the first and third stages is caused by environmental factors and random error factors affecting the pure technical efficiency. (2) The vast majority of cities are in increasing returns to scale, indicating that the improvement of haze control efficiency can be achieved by expanding the scale of haze control inputs.(3) The overall efficiency of haze control in Yangtze River Delta cities from 2014 to 2019 is high and shows a W-shaped change trend with intercity differences. (4) The increase in real GDP per capita and the share of secondary industry have positive effects on improving haze control efficiency, while the population size and urbanization rate have negative effects. Based on this, to further improve the level of haze management efficiency in Yangtze River Delta cities and promote the integrated ecological development of Yangtze River, we should increase technological innovation, create a diversified haze management funding model, and strengthen the level of collaborative management in Yangtze River Delta cities.
In response to the severe haze pollution has become one of the major obstacles to the ecological integration development of the Yangtze River Economic Zone at presenta three-stage DEA model was used to study the haze management efficiency of 27 cities in the Yangtze River Delta from 2014 to 2019. The study shows that (1)Three-stage DEA model can effectively eliminate the effects of environmental factors and random errors, and can truly reflect the level of haze control efficiency in the Yangtze River Delta, and the difference between the integrated efficiency in the first and third stages is caused by environmental factors and random error factors affecting the pure technical efficiency. (2) The vast majority of cities are in increasing returns to scale, indicating that the improvement of haze control efficiency can be achieved by expanding the scale of haze control inputs.(3) The overall efficiency of haze control in Yangtze River Delta cities from 2014 to 2019 is high and shows a W-shaped change trend with intercity differences. (4) The increase in real GDP per capita and the share of secondary industry have positive effects on improving haze control efficiency, while the population size and urbanization rate have negative effects. Based on this, to further improve the level of haze management efficiency in Yangtze River Delta cities and promote the integrated ecological development of Yangtze River, we should increase technological innovation, create a diversified haze management funding model, and strengthen the level of collaborative management in Yangtze River Delta cities.
Abstract:
Rainfall erosivity (R) is related to rainfall amount, rainfall duration, rainfall intensity, and rainfall kinetic energy. It reflects the influence of rainfall characteristics on soil erosion and is one of the important factors affecting soil erosion. Using the daily rainfall data of 25 meteorological stations in Jiangxi Province from 1961 to 2019, based on the rainfall erosivity model, through Mann-Kendall correlation test, wavelet analysis and Kriging spatial interpolation method, the spatial distribution and variation trend of rainfall erosivity in Jiangxi Province were analyzed. The results show that the annual average rainfall and rainfall erosivity in Jiangxi Province gradually increased from southern Jiangxi to northern Jiangxi, and the spatial distribution of rainfall and rainfall erosivity were similar; the rainfall erosivity in spring and summer ranged from 3000 to 6000 MJ·mm· hm-2·h-1·a-1, and the maximum rainfall erosivity in spring is higher than that in summer, while the rainfall erosivity in autumn and winter is significantly smaller than that in spring and summer; the annual rainfall erosivity in Jiangxi Province is evenly distributed, and the average maximum value is distributed in northern Jiangxi area, followed by central Jiangxi and southern Jiangxi, and the erosive force gradually decreased from north to south. The research shows that the erosion factors caused by rainfall are increasing in Jiangxi Province, especially in spring and summer, and there are obvious differences in their spatial distribution.
Rainfall erosivity (R) is related to rainfall amount, rainfall duration, rainfall intensity, and rainfall kinetic energy. It reflects the influence of rainfall characteristics on soil erosion and is one of the important factors affecting soil erosion. Using the daily rainfall data of 25 meteorological stations in Jiangxi Province from 1961 to 2019, based on the rainfall erosivity model, through Mann-Kendall correlation test, wavelet analysis and Kriging spatial interpolation method, the spatial distribution and variation trend of rainfall erosivity in Jiangxi Province were analyzed. The results show that the annual average rainfall and rainfall erosivity in Jiangxi Province gradually increased from southern Jiangxi to northern Jiangxi, and the spatial distribution of rainfall and rainfall erosivity were similar; the rainfall erosivity in spring and summer ranged from 3000 to 6000 MJ·mm· hm-2·h-1·a-1, and the maximum rainfall erosivity in spring is higher than that in summer, while the rainfall erosivity in autumn and winter is significantly smaller than that in spring and summer; the annual rainfall erosivity in Jiangxi Province is evenly distributed, and the average maximum value is distributed in northern Jiangxi area, followed by central Jiangxi and southern Jiangxi, and the erosive force gradually decreased from north to south. The research shows that the erosion factors caused by rainfall are increasing in Jiangxi Province, especially in spring and summer, and there are obvious differences in their spatial distribution.
Abstract:
Nowadays spoofing is the dominant threaten for GNSS receiver due to its low cost and high efficiency. However, the nulling antenna, which is widely used in GNSS receiver, can only mitigate the interference with high power. According to the DOA information of spoofing, some pseudo interferences are added to the received signal and make sure they imping on the antenna array from the same DOAs as those spoofing. Then the following traditional nulling antenna could suppress the spoofing since the beam pattern generate nulls at the DOAs of spoofing. Finally, the simulations of beam pattern and PN code synchronization illustrate the novel performance of proposed methodology.
Nowadays spoofing is the dominant threaten for GNSS receiver due to its low cost and high efficiency. However, the nulling antenna, which is widely used in GNSS receiver, can only mitigate the interference with high power. According to the DOA information of spoofing, some pseudo interferences are added to the received signal and make sure they imping on the antenna array from the same DOAs as those spoofing. Then the following traditional nulling antenna could suppress the spoofing since the beam pattern generate nulls at the DOAs of spoofing. Finally, the simulations of beam pattern and PN code synchronization illustrate the novel performance of proposed methodology.
Abstract:
Due to mission requirements, unmanned craft vehicle(USV) usually operate in the form of formation cooperation and exchange data through wireless ad hoc networks. Due to the influence of ocean waves and other factors, the channel transmission loss of maritime ad hoc networks is usually in a dynamic state. The existing backoff algorithm of MAC protocol in ad hoc networks cannot distinguish between packet collision and packet loss in a dynamic maritime environment, resulting in the decline of reliability and stability. Therefore, this paper proposes an adaptive minimum contention window backoff algorithm based on channel monitoring. The algorithm estimates the channel state by sensing the number of adjacent contention nodes, reduces the channel collision probability and retransmission times, and improves the reliability and stability of the network as a whole. Simulation results show that compared with the classical BEB algorithm, the throughput and fairness of the improved algorithm are increased by 28.67% and 62.00% respectively, and the end-to-end delay and packet loss rate are reduced by 2.84% and 15.10% respectively.
Due to mission requirements, unmanned craft vehicle(USV) usually operate in the form of formation cooperation and exchange data through wireless ad hoc networks. Due to the influence of ocean waves and other factors, the channel transmission loss of maritime ad hoc networks is usually in a dynamic state. The existing backoff algorithm of MAC protocol in ad hoc networks cannot distinguish between packet collision and packet loss in a dynamic maritime environment, resulting in the decline of reliability and stability. Therefore, this paper proposes an adaptive minimum contention window backoff algorithm based on channel monitoring. The algorithm estimates the channel state by sensing the number of adjacent contention nodes, reduces the channel collision probability and retransmission times, and improves the reliability and stability of the network as a whole. Simulation results show that compared with the classical BEB algorithm, the throughput and fairness of the improved algorithm are increased by 28.67% and 62.00% respectively, and the end-to-end delay and packet loss rate are reduced by 2.84% and 15.10% respectively.
Abstract:
In this paper，we consider the transmitter beamformer design at relays for a full-duplex two-way relay system where the two sources are equipped with multiple antennas, and each one of the nearby relays is equipped with two antennas with one for transmission and the other for reception. Under the constraints of the limited transmit power and zero forcing nulling technique used at the relays, the proposed two algorithms are the linear relay beamformers that are respectively to iteratively: minimize the weighted mean square error of the two source nodes and maximize the smaller of signal to noise ratio of the two source nodes The effectiveness and efficiency of the proposed algorithms are evaluated by numerical experiments in terms of the weighted sum rate and the CPU running time.
In this paper，we consider the transmitter beamformer design at relays for a full-duplex two-way relay system where the two sources are equipped with multiple antennas, and each one of the nearby relays is equipped with two antennas with one for transmission and the other for reception. Under the constraints of the limited transmit power and zero forcing nulling technique used at the relays, the proposed two algorithms are the linear relay beamformers that are respectively to iteratively: minimize the weighted mean square error of the two source nodes and maximize the smaller of signal to noise ratio of the two source nodes The effectiveness and efficiency of the proposed algorithms are evaluated by numerical experiments in terms of the weighted sum rate and the CPU running time.
Abstract:
Due to the rapid development of human-computer interaction, affective computing has attracted more and more attention in recent years. Electroencephalogram (EEG) signals are easy to record and difficult to camouflage, it plays an important role in emotion recognition. However, due to the diversity of human emotions and individual variability, EEG-based emotion recognition is still a difficult problem in the field of affective computing. Aiming at this problem, a multi-source domain adaptive dictionary learning and sparse representation approach is proposed in this study. To reduce the difference of data distribution between the source domains and the target domain, the data of all domains are projected into the shared subspace, and a common dictionary is learned in the subspace to establish the relationship between the source domains and the target domain. The learning criterion of common dictionary is to minimize the intra-class sparse reconstruction error and maximize the inter-class sparse reconstruction error, which can ensure that the learned sparse representation has more recognition ability. In addition, to avoid the occurrence of negative transfer, each source domain corresponds to a domain adaptive weight, and the optimal value of the weight can be obtained in the adaptive learning. The model parameters are solved by the method of parameter alternate optimization, and all parameters can reach their optimal solutions at the same time. Experiments on the DEAP dataset verify the effectiveness of our method.
Due to the rapid development of human-computer interaction, affective computing has attracted more and more attention in recent years. Electroencephalogram (EEG) signals are easy to record and difficult to camouflage, it plays an important role in emotion recognition. However, due to the diversity of human emotions and individual variability, EEG-based emotion recognition is still a difficult problem in the field of affective computing. Aiming at this problem, a multi-source domain adaptive dictionary learning and sparse representation approach is proposed in this study. To reduce the difference of data distribution between the source domains and the target domain, the data of all domains are projected into the shared subspace, and a common dictionary is learned in the subspace to establish the relationship between the source domains and the target domain. The learning criterion of common dictionary is to minimize the intra-class sparse reconstruction error and maximize the inter-class sparse reconstruction error, which can ensure that the learned sparse representation has more recognition ability. In addition, to avoid the occurrence of negative transfer, each source domain corresponds to a domain adaptive weight, and the optimal value of the weight can be obtained in the adaptive learning. The model parameters are solved by the method of parameter alternate optimization, and all parameters can reach their optimal solutions at the same time. Experiments on the DEAP dataset verify the effectiveness of our method.
Abstract:
The output characteristics of photovoltaic (PV) array change with the environmental conditions and running state.In order to meet the control requirements of maximum power point tracking (MPPT) under different operation conditions,a sectional control method combining improved quantum particle swarm optimization algorithm（QPSO） and perturb-observe algorithm is proposed after analyzing the output characteristics of photovoltaic array under various working conditions.The inconsistent adaptive mutation DCWQPSO is used to search the maximum power point globally in the initial stage of tracking control to make the power point converge to the maximum power point quickly in order to improve the tracking speed, then the perturb-observe algorithm based on closed-loop fuzzy control is used to search the maximum power point locally to improve the tracking accuracy.The matlab simulation results show that the segmented control method can complete MPPT in only 0.32s under various working conditions of photovoltaic array and remains stable,which has faster tracking speed and higher tracking accuracy than others, so it can improve the efficiency of PV generation effectively.
The output characteristics of photovoltaic (PV) array change with the environmental conditions and running state.In order to meet the control requirements of maximum power point tracking (MPPT) under different operation conditions,a sectional control method combining improved quantum particle swarm optimization algorithm（QPSO） and perturb-observe algorithm is proposed after analyzing the output characteristics of photovoltaic array under various working conditions.The inconsistent adaptive mutation DCWQPSO is used to search the maximum power point globally in the initial stage of tracking control to make the power point converge to the maximum power point quickly in order to improve the tracking speed, then the perturb-observe algorithm based on closed-loop fuzzy control is used to search the maximum power point locally to improve the tracking accuracy.The matlab simulation results show that the segmented control method can complete MPPT in only 0.32s under various working conditions of photovoltaic array and remains stable,which has faster tracking speed and higher tracking accuracy than others, so it can improve the efficiency of PV generation effectively.
Abstract:
High-definition maps (HDM) for autonomous driving (AD) are an important component of AD systems. HDM can provide highly accurate prior data of lane lines and road support facilities for AD systems, but the accuracy evaluation of HDM has been troubled by the development of the industry, so it is an important task to evaluate the accuracy of HDM reasonably. The current methods for relative accuracy evaluation of general maps in the field of mapping are not fully applicable to HDMs. In this study, a method based on point set alignment and resampling is used to evaluate the relative accuracy of lane lines, and experiments are conducted based on relevant real HDM data. That is, the points on the verification curve are fitted and sampled first, then the aligned point pairs are registered, and then resampling is carried out after registration. On this basis, the relative accuracy is calculated. The relative limit error of the verified first group of lane lines is 15.9cm, which is less than 20cm and meets the relative accuracy requirements; the relative limit error of the other three groups of lane lines also meets this requirement. This results show that the relative accuracy of the calculated lane lines based on this method is more accurate and reliable than that of the traditional method, especially the lane lines containing obvious curves. This has a certain influence on the quality control of HDM products, and also provides a theoretical basis and reference for the accuracy evaluation of HDM.
High-definition maps (HDM) for autonomous driving (AD) are an important component of AD systems. HDM can provide highly accurate prior data of lane lines and road support facilities for AD systems, but the accuracy evaluation of HDM has been troubled by the development of the industry, so it is an important task to evaluate the accuracy of HDM reasonably. The current methods for relative accuracy evaluation of general maps in the field of mapping are not fully applicable to HDMs. In this study, a method based on point set alignment and resampling is used to evaluate the relative accuracy of lane lines, and experiments are conducted based on relevant real HDM data. That is, the points on the verification curve are fitted and sampled first, then the aligned point pairs are registered, and then resampling is carried out after registration. On this basis, the relative accuracy is calculated. The relative limit error of the verified first group of lane lines is 15.9cm, which is less than 20cm and meets the relative accuracy requirements; the relative limit error of the other three groups of lane lines also meets this requirement. This results show that the relative accuracy of the calculated lane lines based on this method is more accurate and reliable than that of the traditional method, especially the lane lines containing obvious curves. This has a certain influence on the quality control of HDM products, and also provides a theoretical basis and reference for the accuracy evaluation of HDM.
Abstract:
The non-intrusive load decomposition is to decompose the power signal of a single load device according to the known total power signal. At present, there are many problems based on deep learning models, such as insufficient load feature extraction, low decomposition accuracy, large decomposition error of load equipment with low frequency, and so on. To solve these problems, this paper proposes an attention recurrent neural networks (ARNN) model to realize non-invasive load decomposition, which combines regression network and classification network to solve the non-invasive load decomposition problem. The model extracts the features of sequence signals through RNN network, and uses the attention mechanism to locate the position of important information in the input sequence, so as to improve the representation ability of network. Experiments on public datasets Wiki-Energy and UK-DALE show that our proposed deep neural network is superior to the most advanced neural network under all experimental conditions. We also show that the attention mechanism and auxiliary classification network can correctly detect the opening or closing of devices, and locate the high-power signal part, which improves the accuracy of load decomposition.
The non-intrusive load decomposition is to decompose the power signal of a single load device according to the known total power signal. At present, there are many problems based on deep learning models, such as insufficient load feature extraction, low decomposition accuracy, large decomposition error of load equipment with low frequency, and so on. To solve these problems, this paper proposes an attention recurrent neural networks (ARNN) model to realize non-invasive load decomposition, which combines regression network and classification network to solve the non-invasive load decomposition problem. The model extracts the features of sequence signals through RNN network, and uses the attention mechanism to locate the position of important information in the input sequence, so as to improve the representation ability of network. Experiments on public datasets Wiki-Energy and UK-DALE show that our proposed deep neural network is superior to the most advanced neural network under all experimental conditions. We also show that the attention mechanism and auxiliary classification network can correctly detect the opening or closing of devices, and locate the high-power signal part, which improves the accuracy of load decomposition.
Abstract:
Aiming at the cyber physical system (CPS) under false data injection (FDI) attack, a control method based on sliding mode and extended observer is studied. Firstly, the system is dynamically linearized, an extended observer is constructed, and the convergence condition of the observation error is analyzed. Secondly, the integral sliding mode surface is designed, the asymptotic stability criterion of the sliding mode system is derived by using linear matrix inequality, and the sliding mode vector satisfying the gain performance of the system is obtained. Then, based on the exponential reaching law, an adaptive integral sliding mode controller is proposed to eliminate quantization errors and generalized disturbances, so that the system can reach the sliding surface. The advantages of this method are high estimation accuracy, fast response speed, and strong robustness to FDI attack and quantization parameter mismatch. Finally, numerical simulation verifies the effectiveness of the method.
Aiming at the cyber physical system (CPS) under false data injection (FDI) attack, a control method based on sliding mode and extended observer is studied. Firstly, the system is dynamically linearized, an extended observer is constructed, and the convergence condition of the observation error is analyzed. Secondly, the integral sliding mode surface is designed, the asymptotic stability criterion of the sliding mode system is derived by using linear matrix inequality, and the sliding mode vector satisfying the gain performance of the system is obtained. Then, based on the exponential reaching law, an adaptive integral sliding mode controller is proposed to eliminate quantization errors and generalized disturbances, so that the system can reach the sliding surface. The advantages of this method are high estimation accuracy, fast response speed, and strong robustness to FDI attack and quantization parameter mismatch. Finally, numerical simulation verifies the effectiveness of the method.
Abstract:
The ground temperature changes at a rate of 0.1 K every 10 years, however, due to the influence of solar radiation, when measured in practice, conventional radiation shields sensor will produce a solar radiation error of about 1 K. In order to improve the accuracy of surface temperature measurement and reduced working energy consumption, this paper designs a temperature sensor based on the diffusion of radiant heat from a piezoelectric ceramic vibration acceleration sensor probe. Firstly, the Computational Fluid Dynamics (CFD) method is used to calculate the radiation error of the temperature sensor under multi physics factors, and then the data is fitted and analyzed using the neural network algorithm, and finally the field experiment platform is built to place the temperature sensor in the real environment to verify the feasibility of the scheme. The experimental results show that the absolute error and root mean square error of the correction value and reference value of the surface temperature sensor are 0.041 K and 0.055 K, respectively, which also verifies the superiority of the correction effect of the neural network algorithm.
The ground temperature changes at a rate of 0.1 K every 10 years, however, due to the influence of solar radiation, when measured in practice, conventional radiation shields sensor will produce a solar radiation error of about 1 K. In order to improve the accuracy of surface temperature measurement and reduced working energy consumption, this paper designs a temperature sensor based on the diffusion of radiant heat from a piezoelectric ceramic vibration acceleration sensor probe. Firstly, the Computational Fluid Dynamics (CFD) method is used to calculate the radiation error of the temperature sensor under multi physics factors, and then the data is fitted and analyzed using the neural network algorithm, and finally the field experiment platform is built to place the temperature sensor in the real environment to verify the feasibility of the scheme. The experimental results show that the absolute error and root mean square error of the correction value and reference value of the surface temperature sensor are 0.041 K and 0.055 K, respectively, which also verifies the superiority of the correction effect of the neural network algorithm.
Anti disturbance sliding mode control of fixed wing UAV Based on singular perturbation decomposition
Abstract:
In order to improve the flight control accuracy of fixed wing UAV and reduce the influence of system dynamic coupling and external disturbance on the performance of flight control system of fixed wing UAV. This paper proposed a singular perturbation model of fixed wing UAV and designed a sliding mode control method based on disturbance observer. Firstly, the velocity and attitude of the fixed wing UAV are modeled based on the dynamics of action.Then we transform the dynamic model of fixed wing UAV into singular perturbation model, and then the singular perturbation model is decomposed to complete decoupling. Two reduced order uncoupled subsystems are obtained which is the fast subsystem with angular velocity and the slow subsystem with linear velocity and attitude.The anti disturbance sliding mode controller is designed for angular velocity loop,angle and attitude loop respectively. Finally, the feasibility and effectiveness of the sliding mode control method based on fast and slow decomposition are verified by Simulink simulation.
In order to improve the flight control accuracy of fixed wing UAV and reduce the influence of system dynamic coupling and external disturbance on the performance of flight control system of fixed wing UAV. This paper proposed a singular perturbation model of fixed wing UAV and designed a sliding mode control method based on disturbance observer. Firstly, the velocity and attitude of the fixed wing UAV are modeled based on the dynamics of action.Then we transform the dynamic model of fixed wing UAV into singular perturbation model, and then the singular perturbation model is decomposed to complete decoupling. Two reduced order uncoupled subsystems are obtained which is the fast subsystem with angular velocity and the slow subsystem with linear velocity and attitude.The anti disturbance sliding mode controller is designed for angular velocity loop,angle and attitude loop respectively. Finally, the feasibility and effectiveness of the sliding mode control method based on fast and slow decomposition are verified by Simulink simulation.
Abstract:
Aiming at the problem of low efficiency and accuracy of fireworks detection due to the small size of fire target and the confusion of fire feature with actual scene in complex environment, a small scale fireworks target detection method based on improved YOLOv5 is proposed. Firstly, a fourth detection layer is added to the third detection layer output in the original YOLOv5 model, so as to obtain a larger feature map for small target detection and strengthen the feature extraction capability of the network model. Secondly, in order to solve the problem that the target is prone to miss detection in the shielded scene, GIOU_Loss used to calculate the regression loss function of the target frame in the original network is replaced by DIOU_Loss. Finally, TensorRT is used to compress and accelerate the optimization of the model, and it is deployed to the Jetson TX2 development board for accelerated reasoning experiments. More fireworks scene data are constructed by replication enhancement method and a large number of experimental results show that the proposed method not only has a fast convergence speed, but also has a higher accuracy for small scale fire spot detection. It is suitable for popularization and application.
Aiming at the problem of low efficiency and accuracy of fireworks detection due to the small size of fire target and the confusion of fire feature with actual scene in complex environment, a small scale fireworks target detection method based on improved YOLOv5 is proposed. Firstly, a fourth detection layer is added to the third detection layer output in the original YOLOv5 model, so as to obtain a larger feature map for small target detection and strengthen the feature extraction capability of the network model. Secondly, in order to solve the problem that the target is prone to miss detection in the shielded scene, GIOU_Loss used to calculate the regression loss function of the target frame in the original network is replaced by DIOU_Loss. Finally, TensorRT is used to compress and accelerate the optimization of the model, and it is deployed to the Jetson TX2 development board for accelerated reasoning experiments. More fireworks scene data are constructed by replication enhancement method and a large number of experimental results show that the proposed method not only has a fast convergence speed, but also has a higher accuracy for small scale fire spot detection. It is suitable for popularization and application.
Abstract:
In this paper, the robust control problem of singular Markovian jump system with parametric uncertainties and time-varying delays is studied. By constructing the delay-dependent Lyapunov-Krasovskii function and combining with the inequality scaling lemma, the LMI condition of the stochastic tolerance is given which meanwhile satisfies the performance index of the system, and the state feedback controller satisfying the condition is designed by using the variable substitution method.
In this paper, the robust control problem of singular Markovian jump system with parametric uncertainties and time-varying delays is studied. By constructing the delay-dependent Lyapunov-Krasovskii function and combining with the inequality scaling lemma, the LMI condition of the stochastic tolerance is given which meanwhile satisfies the performance index of the system, and the state feedback controller satisfying the condition is designed by using the variable substitution method.
Abstract:
In order to solve the problems of color distortion, key information blur and detail loss of underwater image, an underwater image enhancement method based on SK attention residual network is proposed. The generator structure in the generative adversarial network is improved, and residual module is introduced to reduce the feature loss between the encoder and decoder, and enhance the image detail and color. In order to make the network adapt to different scale feature maps to extract key information of images, SK attention mechanism is added after the residual module. At the same time,a parametric rectified linear unit is used to improve the fitting ability of the network. This method is verified in real and synthetic underwater image datasets, and the traditional method and deep learning method are used for subjective and objective evaluation. In the subjective effect analysis, it is found that the color, key information and detail features of the enhanced image have been greatly improved. In the objective evaluation index, it is found that the index values of this method are higher than the existing underwater image enhancement algorithms, which shows the effectiveness of this algorithm.
In order to solve the problems of color distortion, key information blur and detail loss of underwater image, an underwater image enhancement method based on SK attention residual network is proposed. The generator structure in the generative adversarial network is improved, and residual module is introduced to reduce the feature loss between the encoder and decoder, and enhance the image detail and color. In order to make the network adapt to different scale feature maps to extract key information of images, SK attention mechanism is added after the residual module. At the same time,a parametric rectified linear unit is used to improve the fitting ability of the network. This method is verified in real and synthetic underwater image datasets, and the traditional method and deep learning method are used for subjective and objective evaluation. In the subjective effect analysis, it is found that the color, key information and detail features of the enhanced image have been greatly improved. In the objective evaluation index, it is found that the index values of this method are higher than the existing underwater image enhancement algorithms, which shows the effectiveness of this algorithm.
Abstract:
In order to timely identify the changing trend of marine environment and reduce the influence of long-term accumulated marine environment data on the prediction model, an online prediction model of marine environment data based on cyclic online sequential extreme learning machine was proposed. The marine environment data training set is initialized by online method, the existing marine environment data is input block by block by online sequential extreme learning machine algorithm, and the input weight is cyclically processed by automatic coding technology of extreme learning machine and a normalized method, and the prediction model is updated online. Finally, online prediction of marine environmental data is completed. The model was used to predict dissolved oxygen, chlorophyll A, turbidity, and blue-green algae. The results show that the prediction accuracy of R-OSELM model is better than that of the comparison model. It is confirmed that R-OSELM model has the online prediction ability of marine environmental data, which can provide reference for marine eutrophication and marine environmental pollution early warning.
In order to timely identify the changing trend of marine environment and reduce the influence of long-term accumulated marine environment data on the prediction model, an online prediction model of marine environment data based on cyclic online sequential extreme learning machine was proposed. The marine environment data training set is initialized by online method, the existing marine environment data is input block by block by online sequential extreme learning machine algorithm, and the input weight is cyclically processed by automatic coding technology of extreme learning machine and a normalized method, and the prediction model is updated online. Finally, online prediction of marine environmental data is completed. The model was used to predict dissolved oxygen, chlorophyll A, turbidity, and blue-green algae. The results show that the prediction accuracy of R-OSELM model is better than that of the comparison model. It is confirmed that R-OSELM model has the online prediction ability of marine environmental data, which can provide reference for marine eutrophication and marine environmental pollution early warning.
Abstract:
Abstract: Indices such as tumor cell density, nucleocytoplasmic ratio, and average size have important implications for cancer grading and prognosis. Therefore, nuclear segmentation is the basis of tumor microenvironment analysis in computational pathology. In addition, the exploration of new tumor markers is of great significance through statistical analysis of segmentation results. However, the morphology of nuclei in the background of pathological images is irregular, the staining of nuclei is uneven, and there is a problem with adhesion between the edges of nuclei. In the existing deep learning algorithm, when the main body of the nucleus is correctly segmented, the segmentation error of the edge will not have much impact on the overall loss, so the adhering nucleus can easily be regarded as the same segmentation target. In order to solve the nuclear overlapping problems, a new segmentation algorithm based on the Transformer and Distance map, TDM-Net, is proposed, which integrates the core of multi-head self-attention mechanism in Transformer with contextual information to fully explore the proximity relationship and enhances the learning ability of image details by introducing distance map to emphasize the interior of nuclei and weaken the boundary of nuclei. The algorithm's dice coefficient, precision, aggregated jaccard index and hausdorff distance are 0.7979, 0.7561, 0.6672, and 10.11, respectively. The results show that TDM-Net performs better than other segmentation algorithms, effectively improving nuclei segmentation accuracy and solving overlapping problems with different nuclei.
Abstract: Indices such as tumor cell density, nucleocytoplasmic ratio, and average size have important implications for cancer grading and prognosis. Therefore, nuclear segmentation is the basis of tumor microenvironment analysis in computational pathology. In addition, the exploration of new tumor markers is of great significance through statistical analysis of segmentation results. However, the morphology of nuclei in the background of pathological images is irregular, the staining of nuclei is uneven, and there is a problem with adhesion between the edges of nuclei. In the existing deep learning algorithm, when the main body of the nucleus is correctly segmented, the segmentation error of the edge will not have much impact on the overall loss, so the adhering nucleus can easily be regarded as the same segmentation target. In order to solve the nuclear overlapping problems, a new segmentation algorithm based on the Transformer and Distance map, TDM-Net, is proposed, which integrates the core of multi-head self-attention mechanism in Transformer with contextual information to fully explore the proximity relationship and enhances the learning ability of image details by introducing distance map to emphasize the interior of nuclei and weaken the boundary of nuclei. The algorithm's dice coefficient, precision, aggregated jaccard index and hausdorff distance are 0.7979, 0.7561, 0.6672, and 10.11, respectively. The results show that TDM-Net performs better than other segmentation algorithms, effectively improving nuclei segmentation accuracy and solving overlapping problems with different nuclei.
Abstract:
In order to improve the intelligent level and security of mobile robot autonomous navigation system, an autonomous navigation system based on Soft Actor-Critic algorithm under the security barrier mechanism was pro-posed to make the robot more intelligent. The return function based on the distance between the robot and the nearest obstacle, the distance from the target point and the yaw angle was designed. In the gazebo simulation platform, a mobile robot with lidar and its surrounding environment were built. Experiments showed that the security barrier mechanism reduced the probability of robot hitting obstacles to a certain extent, improved the success rate of navigation, and made the mobile robot autonomous navigation system based on Soft Actor-Critic algorithm have higher generalization ability. The system still had the ability of autonomous navigation when changing the origin and destination or even changing the static environment to dynamic.
In order to improve the intelligent level and security of mobile robot autonomous navigation system, an autonomous navigation system based on Soft Actor-Critic algorithm under the security barrier mechanism was pro-posed to make the robot more intelligent. The return function based on the distance between the robot and the nearest obstacle, the distance from the target point and the yaw angle was designed. In the gazebo simulation platform, a mobile robot with lidar and its surrounding environment were built. Experiments showed that the security barrier mechanism reduced the probability of robot hitting obstacles to a certain extent, improved the success rate of navigation, and made the mobile robot autonomous navigation system based on Soft Actor-Critic algorithm have higher generalization ability. The system still had the ability of autonomous navigation when changing the origin and destination or even changing the static environment to dynamic.
Abstract:
Global climate change caused by excessive greenhouse gas (GHGs) emissions has been widely concerned. Agricultural activities are the second largest source of GHGs emissions, it is urgent to reduce agricultural GHGs emissions. Biochar, which has stable properties, abundant aromatic carbon and pores, is produced by pyrolysis of biomass under high temperature and limited oxygen conditions. The effect of biochar amendment on GHGs mitigation and soil carbon sequestration is excellent, and biochar application has the potential to participate in China’s ongoing carbon trading of voluntary emission reduction (VER). However, the factors affecting the carbon sequestration and GHGs emission reduction effect of biochar are complicated, so it is necessary to systematically summarize the mitigation effect, influencing factors and research progress of biochar. This paper reviewed the GHGs emission reduction and carbon sequestration effect of biochar through pot and field experiment as well as meta-analysis research. At the same time, Citespace software was used for visual analysis to explore the research hotspot and development trend in this field. The opportunities and challenges faced by biochar application projects participating in carbon trading were summarized based on the characteristics of domestic and foreign carbon trading market development and corresponding supporting policies. Corresponding solutions were provided also, which offered scientific guidance and useful reference for the development of carbon sequestration and GHGs emission reduction research of biochar and the successful participation of biochar application projects in carbon trading.
Global climate change caused by excessive greenhouse gas (GHGs) emissions has been widely concerned. Agricultural activities are the second largest source of GHGs emissions, it is urgent to reduce agricultural GHGs emissions. Biochar, which has stable properties, abundant aromatic carbon and pores, is produced by pyrolysis of biomass under high temperature and limited oxygen conditions. The effect of biochar amendment on GHGs mitigation and soil carbon sequestration is excellent, and biochar application has the potential to participate in China’s ongoing carbon trading of voluntary emission reduction (VER). However, the factors affecting the carbon sequestration and GHGs emission reduction effect of biochar are complicated, so it is necessary to systematically summarize the mitigation effect, influencing factors and research progress of biochar. This paper reviewed the GHGs emission reduction and carbon sequestration effect of biochar through pot and field experiment as well as meta-analysis research. At the same time, Citespace software was used for visual analysis to explore the research hotspot and development trend in this field. The opportunities and challenges faced by biochar application projects participating in carbon trading were summarized based on the characteristics of domestic and foreign carbon trading market development and corresponding supporting policies. Corresponding solutions were provided also, which offered scientific guidance and useful reference for the development of carbon sequestration and GHGs emission reduction research of biochar and the successful participation of biochar application projects in carbon trading.
Abstract:
In order to prevent the mosaic line from passing through the visually significant features when the orthophoto map is mosaic, and damage the integrity of the features in the image. In this paper, a mosaic line extraction method for orthophoto images is proposed, which combines edge information optimization LSC and improved A-star algorithm. Firstly, the super-pixel segmentation theory of linear spectral clustering algorithm (LSC) is introduced into the mosaic line extraction algorithm of orthophoto images. The classical linear spectral clustering algorithm is optimized by using edge intensity factor to effectively use the spectral information and edge information in orthophoto images. Secondly, the improved linear spectral clustering algorithm is applied to the overlapping areas of two orthophoto images to obtain the boundary feature maps of various ground objects; Then, the edge irregularity and isolated noise in the boundary feature map are removed by mathematical morphology; Finally, the improved A-star algorithm is used to search the shortest path in the boundary map to quickly obtain the optimal image mosaic line. This method is compared with related methods by using the aerial orthophoto Image of a real UAV, The results show that the proposed method can extract mosaic lines efficiently and with high quality. It effectively bypasses the visual salient features and meets the application requirements of the actual orthophoto map making.
In order to prevent the mosaic line from passing through the visually significant features when the orthophoto map is mosaic, and damage the integrity of the features in the image. In this paper, a mosaic line extraction method for orthophoto images is proposed, which combines edge information optimization LSC and improved A-star algorithm. Firstly, the super-pixel segmentation theory of linear spectral clustering algorithm (LSC) is introduced into the mosaic line extraction algorithm of orthophoto images. The classical linear spectral clustering algorithm is optimized by using edge intensity factor to effectively use the spectral information and edge information in orthophoto images. Secondly, the improved linear spectral clustering algorithm is applied to the overlapping areas of two orthophoto images to obtain the boundary feature maps of various ground objects; Then, the edge irregularity and isolated noise in the boundary feature map are removed by mathematical morphology; Finally, the improved A-star algorithm is used to search the shortest path in the boundary map to quickly obtain the optimal image mosaic line. This method is compared with related methods by using the aerial orthophoto Image of a real UAV, The results show that the proposed method can extract mosaic lines efficiently and with high quality. It effectively bypasses the visual salient features and meets the application requirements of the actual orthophoto map making.
Abstract:
In order to explore the illumination adaptability of environmental perception equipment in the application of SLAM algorithm, verification evaluation experiments of lidar and depth camera SLAM algorithms were carried out under different illumination intensities. Based on a four-wheel differential robot, equipped with a 16-line lidar and a depth camera, combined with the LOAM(Lidar Odometry And Mapping) and RTAB_MAP (Real-Time Appearance-Based Mapping) algorithms, the lighting adaptability of the device is analyzed and verified in light and dark environments respectively. The experimental results show that in a bright environment, the median errors of the Visual SLAM and LiDAR SLAM system deviations are 0.203m and 0.644m, respectively; in the dark environment, the deviations of the two are 0.282m and 0.683m, respectively; The positioning and mapping effect of the depth camera in both bright and dark environments is better than that of the lidar, and the depth camera is more adaptable.
In order to explore the illumination adaptability of environmental perception equipment in the application of SLAM algorithm, verification evaluation experiments of lidar and depth camera SLAM algorithms were carried out under different illumination intensities. Based on a four-wheel differential robot, equipped with a 16-line lidar and a depth camera, combined with the LOAM(Lidar Odometry And Mapping) and RTAB_MAP (Real-Time Appearance-Based Mapping) algorithms, the lighting adaptability of the device is analyzed and verified in light and dark environments respectively. The experimental results show that in a bright environment, the median errors of the Visual SLAM and LiDAR SLAM system deviations are 0.203m and 0.644m, respectively; in the dark environment, the deviations of the two are 0.282m and 0.683m, respectively; The positioning and mapping effect of the depth camera in both bright and dark environments is better than that of the lidar, and the depth camera is more adaptable.
Abstract:
In order to solve the problems of high school operating costs and poor school bus service quality due to the scattered distribution of bus stops in rural areas, a multi-objective SBRP model was developed for the mixed-load and no mixed-load scenarios. In the no mixed-load scenario, a model of the SBRP is developed to optimize the student travel cost and school operation cost, while in the mixed-load scenario, a model of the SBRP is developed to consider the input cost and operation cost of the school bus. Tests were conducted on an international benchmark case. Several heuristic algorithms were compared, and the annealing algorithms were selected to solve the models. Different search operators were introduced to solve the models constructed under different scenarios, and the experimental results proved the efficacy of the constructed models, which can improve school bus quality of service and reduce operating costs in rural areas.
In order to solve the problems of high school operating costs and poor school bus service quality due to the scattered distribution of bus stops in rural areas, a multi-objective SBRP model was developed for the mixed-load and no mixed-load scenarios. In the no mixed-load scenario, a model of the SBRP is developed to optimize the student travel cost and school operation cost, while in the mixed-load scenario, a model of the SBRP is developed to consider the input cost and operation cost of the school bus. Tests were conducted on an international benchmark case. Several heuristic algorithms were compared, and the annealing algorithms were selected to solve the models. Different search operators were introduced to solve the models constructed under different scenarios, and the experimental results proved the efficacy of the constructed models, which can improve school bus quality of service and reduce operating costs in rural areas.
Abstract:
With the ageing of the original railway and the increase in demand for capacity, the existing railway lines need to be altered, so the existing lines need to be surveyed without affecting normal operations. At present, most of the existing railway lines are surveyed by traditional manual measurement methods, which are inefficient, heavy, cumbersome and repeatedly on line. To address the above shortcomings, this paper uses a least squares based 2D template point cloud matching algorithm to jointly and repeatedly calculate the rail track surface centre point position and obtain the track centre line and track gauge and other related parameters based on the absolute 3D point cloud data obtained by the track mobile laser scanning system. The test and result analysis at the track test site show that the algorithm can effectively extract the track centre line and provide accurate centre line data for the existing line survey.
With the ageing of the original railway and the increase in demand for capacity, the existing railway lines need to be altered, so the existing lines need to be surveyed without affecting normal operations. At present, most of the existing railway lines are surveyed by traditional manual measurement methods, which are inefficient, heavy, cumbersome and repeatedly on line. To address the above shortcomings, this paper uses a least squares based 2D template point cloud matching algorithm to jointly and repeatedly calculate the rail track surface centre point position and obtain the track centre line and track gauge and other related parameters based on the absolute 3D point cloud data obtained by the track mobile laser scanning system. The test and result analysis at the track test site show that the algorithm can effectively extract the track centre line and provide accurate centre line data for the existing line survey.
Abstract:
In order to reduce the sulfur and olefin content and the loss of octane number, to ensure the clean production of gasoline. Based on the data accumulated by the operation of the S Zorb device, first use Lasso to initially screen the modeling variables, and calculate the index factor contribution based on the BP neural network, and further screen out 15 main variables to build the octane loss prediction model. Secondly, the four models are compared and analyzed, and it is concluded that the BP neural network has better prediction accuracy and is more suitable as an octane loss prediction model. After ten-fold cross-validation, the average MSE value is 0.027193 and the average R2 value is 0.90487, which verifies the reliability of the model. sex. Finally, under the premise that the sulfur content of the oil is not greater than 5μg/g, the main variables are optimized and controlled in combination with multiple linear regression. The results show that it is necessary to change multiple variables at the same time to reduce the octane loss by more than 30%. The multiple linear regression model has good prediction accuracy and can regulate the main variables in a certain proportion. The trajectory of the alkane number and sulfur content.
In order to reduce the sulfur and olefin content and the loss of octane number, to ensure the clean production of gasoline. Based on the data accumulated by the operation of the S Zorb device, first use Lasso to initially screen the modeling variables, and calculate the index factor contribution based on the BP neural network, and further screen out 15 main variables to build the octane loss prediction model. Secondly, the four models are compared and analyzed, and it is concluded that the BP neural network has better prediction accuracy and is more suitable as an octane loss prediction model. After ten-fold cross-validation, the average MSE value is 0.027193 and the average R2 value is 0.90487, which verifies the reliability of the model. sex. Finally, under the premise that the sulfur content of the oil is not greater than 5μg/g, the main variables are optimized and controlled in combination with multiple linear regression. The results show that it is necessary to change multiple variables at the same time to reduce the octane loss by more than 30%. The multiple linear regression model has good prediction accuracy and can regulate the main variables in a certain proportion. The trajectory of the alkane number and sulfur content.
Abstract:
The water level change in Poyang Lake is very complex, while traditional techniques are difficult to monitor such dynamic changes with large scale and high precision. In September 2018, the Ice, Cloud and Land Elevation 2 (ICESat-2) satellite, equipped with the world's advanced photon counting laser altimeter, was successfully launched, which provides a good opportunity to monitor the dynamic changes of water levels in medium and large inland lakes. In this study, the monthly ATL13 global inland water data product of ICESat-2 satellite from 2018 to 2020 are used to estimate and analyze the water level variation of Poyang Lake. The measured data of Hukou, Xingzi and Kangshan hydrology stations were used for verification and error correction, and the water level and rainfall data of each station were combined to analyze the dynamic variation of Poyang Lake water level and its causes. The results showed that: (1) The annual water level of Poyang Lake varied sharply with obvious seasonal variation, and the overall water level showed an increasing trend. The high-water level period was from June to October, and the low water level period was in other months, and the peak of water level appeared from July to September. (2) ICESat-2 has a high linear correlation with the measured water level with the coefficient of above 0.846, and the coefficient is as high as 0.974 after error correction. The root mean square error (RMSE) is 1.660m at Hukou station, 1.073m at Xingzi station, and 0.836m at Kangshan station. After error correction and recalcitration, the measurement accuracy is improved by nearly 1m, and the RMSE is 0.663m at Hukou station, 0.659m at Xingzi station, and 0.440 m at Kangshan station. (3) The variation of water level in Poyang Lake is highly correlated with the variation of rainfall. The precipitation is low in January-February and October-December, and the water level decreases in the dry season. From March to October, the rainfall increases, and the water level rises. Rainfall mainly occurs from July to September, when is the peak of water level.
The water level change in Poyang Lake is very complex, while traditional techniques are difficult to monitor such dynamic changes with large scale and high precision. In September 2018, the Ice, Cloud and Land Elevation 2 (ICESat-2) satellite, equipped with the world's advanced photon counting laser altimeter, was successfully launched, which provides a good opportunity to monitor the dynamic changes of water levels in medium and large inland lakes. In this study, the monthly ATL13 global inland water data product of ICESat-2 satellite from 2018 to 2020 are used to estimate and analyze the water level variation of Poyang Lake. The measured data of Hukou, Xingzi and Kangshan hydrology stations were used for verification and error correction, and the water level and rainfall data of each station were combined to analyze the dynamic variation of Poyang Lake water level and its causes. The results showed that: (1) The annual water level of Poyang Lake varied sharply with obvious seasonal variation, and the overall water level showed an increasing trend. The high-water level period was from June to October, and the low water level period was in other months, and the peak of water level appeared from July to September. (2) ICESat-2 has a high linear correlation with the measured water level with the coefficient of above 0.846, and the coefficient is as high as 0.974 after error correction. The root mean square error (RMSE) is 1.660m at Hukou station, 1.073m at Xingzi station, and 0.836m at Kangshan station. After error correction and recalcitration, the measurement accuracy is improved by nearly 1m, and the RMSE is 0.663m at Hukou station, 0.659m at Xingzi station, and 0.440 m at Kangshan station. (3) The variation of water level in Poyang Lake is highly correlated with the variation of rainfall. The precipitation is low in January-February and October-December, and the water level decreases in the dry season. From March to October, the rainfall increases, and the water level rises. Rainfall mainly occurs from July to September, when is the peak of water level.
Abstract:
In order to improve the classification performance of hyperspectral images with limited training samples, a hyperspectral image classification network (DPAMN) based on double pooling attention mechanism is proposed in this paper. Firstly, DPAMN uses three-dimensional convolution to extract the spatial and spectral shallow information of hyperspectral images. Secondly, in order to enhance the feature extraction ability of the network, a double pooling attention mechanism is introduced into DPAMN. Finally, the three-dimensional convolution dense connection module is introduced into the deep layer of the network, which can not only fully extract the spatial and spectral features of hyperspectral images, but also improve the ability of feature discrimination. Experiments show that the overall average accuracy of 95.45%, 97.11%, 95.30% and 93.71% can be achieved on Indian pines, University of Pavia, Salinas and Houston 2013 data sets. Compared with the current mainstream existing advanced methods, the method proposed in this paper is greatly improved on the four data sets, which shows that the proposed method has strong generalization ability.
In order to improve the classification performance of hyperspectral images with limited training samples, a hyperspectral image classification network (DPAMN) based on double pooling attention mechanism is proposed in this paper. Firstly, DPAMN uses three-dimensional convolution to extract the spatial and spectral shallow information of hyperspectral images. Secondly, in order to enhance the feature extraction ability of the network, a double pooling attention mechanism is introduced into DPAMN. Finally, the three-dimensional convolution dense connection module is introduced into the deep layer of the network, which can not only fully extract the spatial and spectral features of hyperspectral images, but also improve the ability of feature discrimination. Experiments show that the overall average accuracy of 95.45%, 97.11%, 95.30% and 93.71% can be achieved on Indian pines, University of Pavia, Salinas and Houston 2013 data sets. Compared with the current mainstream existing advanced methods, the method proposed in this paper is greatly improved on the four data sets, which shows that the proposed method has strong generalization ability.
Abstract:
Day-ahead load forecasting is an important task for the power dispatching center to formulate dispatching plans. It is of great significance for formulating reasonable dispatching plans and ensuring the safety and reliability of power system operation. The time series of power loads experience random errors. The structures of intelligent algorithm based prediction models are complex, and calculation loads are heavy to fully extract load information. To solve this problem, this paper proposes a day-ahead power load forecasting method based on repeated wavelet transform-support vector machine（RWT-SVM） by using the historical power load time series of distributed power grids. The method uses WT to decompose the power load time series of distributed power grids into multiple subsequences; the MAE is applied to calculate the prediction error contribution of each subsequence; further decomposes the sequence with the largest MAE to improve the prediction ability of the model. The simulation results show that the forecasting method based on RWT-SVM is more accurate than other methods.
Day-ahead load forecasting is an important task for the power dispatching center to formulate dispatching plans. It is of great significance for formulating reasonable dispatching plans and ensuring the safety and reliability of power system operation. The time series of power loads experience random errors. The structures of intelligent algorithm based prediction models are complex, and calculation loads are heavy to fully extract load information. To solve this problem, this paper proposes a day-ahead power load forecasting method based on repeated wavelet transform-support vector machine（RWT-SVM） by using the historical power load time series of distributed power grids. The method uses WT to decompose the power load time series of distributed power grids into multiple subsequences; the MAE is applied to calculate the prediction error contribution of each subsequence; further decomposes the sequence with the largest MAE to improve the prediction ability of the model. The simulation results show that the forecasting method based on RWT-SVM is more accurate than other methods.
Abstract:
In view of the different importance of input load characteristics to the decomposition results and the high decomposition error caused by the limited time dependence of LSTM in capturing long-term power consumption information, a non-intrusive load decomposition model based on multi-attention mechanism integration is proposed. Firstly, the probsparse self-attention mechanism is used to optimize the load characteristics extracted by one-dimensional dilated convolution. Then, the temporal pattern attention mechanism is used to give weight to the hidden state of LSTM, so as to enhance the learning ability of the network on the time dependence of long-term power consumption information. Finally, the validity of the proposed decomposition model is verified using the publicly available dataset UKDALE and REDD. Experimental results show that, compared with other decomposition algorithms, the decomposition algorithm based on multi-attention mechanism integration not only has the ability to select important load features, but also can correctly establish the time-dependent relationship between features and effectively reduce the decomposition error.
In view of the different importance of input load characteristics to the decomposition results and the high decomposition error caused by the limited time dependence of LSTM in capturing long-term power consumption information, a non-intrusive load decomposition model based on multi-attention mechanism integration is proposed. Firstly, the probsparse self-attention mechanism is used to optimize the load characteristics extracted by one-dimensional dilated convolution. Then, the temporal pattern attention mechanism is used to give weight to the hidden state of LSTM, so as to enhance the learning ability of the network on the time dependence of long-term power consumption information. Finally, the validity of the proposed decomposition model is verified using the publicly available dataset UKDALE and REDD. Experimental results show that, compared with other decomposition algorithms, the decomposition algorithm based on multi-attention mechanism integration not only has the ability to select important load features, but also can correctly establish the time-dependent relationship between features and effectively reduce the decomposition error.
Abstract:
In the past few years, robots have become an important means of substation inspection, and robotic inspection technology for non-fixed lines has received increasing attention in order to perform inspection tasks more flexibly. How to achieve high-precision positioning in complex substation environments is one of the core problems to be solved. It is difficult for a single sensor to meet the requirements of reliable positioning in substations, therefore, this paper designs a multi-sensor fusion LINS-GNSS positioning method. Its front-end tightly couples LiDAR and inertial navigation based on an iterative error-state Kalman filter framework, which recursively corrects the estimated state by generating new feature correspondences in each iteration. The back-end uses a factor graph optimization approach to loosely couple the localization results from the satellite navigation with the localization results output from the LiDAR-SLAM back-end. The optimization process first aligns the local coordinate system with the global coordinate system, then adds the position constraints of the GNSS as a priori edge to the factor graph in the back-end, and finally outputs the positioning results in the global coordinate system. In order to evaluate the performance of the LINS-GNSS system in the substation environment, this paper conducted field tests of real scenarios. The experimental results show that the LINS-GNSS system can achieve a positioning accuracy better than 1m in the substation environment.
In the past few years, robots have become an important means of substation inspection, and robotic inspection technology for non-fixed lines has received increasing attention in order to perform inspection tasks more flexibly. How to achieve high-precision positioning in complex substation environments is one of the core problems to be solved. It is difficult for a single sensor to meet the requirements of reliable positioning in substations, therefore, this paper designs a multi-sensor fusion LINS-GNSS positioning method. Its front-end tightly couples LiDAR and inertial navigation based on an iterative error-state Kalman filter framework, which recursively corrects the estimated state by generating new feature correspondences in each iteration. The back-end uses a factor graph optimization approach to loosely couple the localization results from the satellite navigation with the localization results output from the LiDAR-SLAM back-end. The optimization process first aligns the local coordinate system with the global coordinate system, then adds the position constraints of the GNSS as a priori edge to the factor graph in the back-end, and finally outputs the positioning results in the global coordinate system. In order to evaluate the performance of the LINS-GNSS system in the substation environment, this paper conducted field tests of real scenarios. The experimental results show that the LINS-GNSS system can achieve a positioning accuracy better than 1m in the substation environment.
Abstract:
The traditional security extraction technology of big data concealment features ignores the public key and key encapsulation of big data ciphertext, and the category of big data concealment features is chaotic, leading to the low extraction accuracy and high redundancy of this technology. Therefore, this paper proposes a security extraction method of big data concealment features based on mixed cipher system. Big data ciphertext is generated by public key encapsulation and key encapsulation mechanism in mixed cryptography. Symmetric encryption method and asymmetric encryption method are designed according to the ciphertext content, based on the hidden characteristics of the classification. The hidden feature phase space of big data is constructed by using the hidden features of different classes, and the correlation dimension value among big data is calculated to realize the safe extraction of the hidden features of big data. The experimental results show that, compared with the traditional methods, the proposed big data hidden feature extraction method has low redundancy, the average correct classification rate of big data hidden features is up to 95%, and the error of feature safe extraction is low, which verifies that the proposed method has better application performance.
The traditional security extraction technology of big data concealment features ignores the public key and key encapsulation of big data ciphertext, and the category of big data concealment features is chaotic, leading to the low extraction accuracy and high redundancy of this technology. Therefore, this paper proposes a security extraction method of big data concealment features based on mixed cipher system. Big data ciphertext is generated by public key encapsulation and key encapsulation mechanism in mixed cryptography. Symmetric encryption method and asymmetric encryption method are designed according to the ciphertext content, based on the hidden characteristics of the classification. The hidden feature phase space of big data is constructed by using the hidden features of different classes, and the correlation dimension value among big data is calculated to realize the safe extraction of the hidden features of big data. The experimental results show that, compared with the traditional methods, the proposed big data hidden feature extraction method has low redundancy, the average correct classification rate of big data hidden features is up to 95%, and the error of feature safe extraction is low, which verifies that the proposed method has better application performance.
DONG Pingxian, GUO Fang, CHEN Cheng, SONG Xiaofan, WANG Hui, BAI Pingping, QI Huanruo, QIAN Yiming, ZHANG Haojie, HAN Yunhao
Abstract:
Aiming at the problems of large error and low efficiency in cable laying based on traditional manual design, the computer-aided design optimized by ant colony algorithm is applied to cable laying path planning. The ant colony algorithm (ACA) is used to solve the shortest path problem of cable laying by taking advantage of the advantages of multi terminal line calculation in complex lines. At the same time, this paper planarizes the cable laying path, and further optimizes it by using Gompertz function from two aspects: pheromone restriction and self-adaptive adjustment of volatilization factor, which improves the convergence speed and global performance of ant colony algorithm. The simulation results show that the optimized ant colony algorithm can quickly get the shortest cable laying path in the process of substation digital three-dimensional cable laying, save the cost of manpower and materials, and improve the design accuracy.
Aiming at the problems of large error and low efficiency in cable laying based on traditional manual design, the computer-aided design optimized by ant colony algorithm is applied to cable laying path planning. The ant colony algorithm (ACA) is used to solve the shortest path problem of cable laying by taking advantage of the advantages of multi terminal line calculation in complex lines. At the same time, this paper planarizes the cable laying path, and further optimizes it by using Gompertz function from two aspects: pheromone restriction and self-adaptive adjustment of volatilization factor, which improves the convergence speed and global performance of ant colony algorithm. The simulation results show that the optimized ant colony algorithm can quickly get the shortest cable laying path in the process of substation digital three-dimensional cable laying, save the cost of manpower and materials, and improve the design accuracy.
Abstract:
Integrated energy system including wind power, photovoltaic and other new energy to achieve electricity, heat, cold and other complementary energy supply has attracted much attention. When multiple investors participate in the operation of integrated energy system as independent subjects, it is worth paying attention to how to rationally allocate the capacity of each equipment to better absorb new energy and maximize the interests of each investor. Based on the Nash equilibrium principle of game theory, this paper establishes a capacity allocation game model for the comprehensive energy system composed of wind power equipment, photovoltaic equipment and cogeneration equipment, and uses particle swarm optimization algorithm to solve it. The comparative analysis of the three different scenarios of non-game, non-cooperative game and cooperative game shows that in the cooperative game scenario, the investor's return is relatively the largest, the capacity allocation is relatively the smallest, the system has the largest overall return, and each participant has the obvious possibility of cooperation. This provides a solution for multi-party participation in the energy supply market.
Integrated energy system including wind power, photovoltaic and other new energy to achieve electricity, heat, cold and other complementary energy supply has attracted much attention. When multiple investors participate in the operation of integrated energy system as independent subjects, it is worth paying attention to how to rationally allocate the capacity of each equipment to better absorb new energy and maximize the interests of each investor. Based on the Nash equilibrium principle of game theory, this paper establishes a capacity allocation game model for the comprehensive energy system composed of wind power equipment, photovoltaic equipment and cogeneration equipment, and uses particle swarm optimization algorithm to solve it. The comparative analysis of the three different scenarios of non-game, non-cooperative game and cooperative game shows that in the cooperative game scenario, the investor's return is relatively the largest, the capacity allocation is relatively the smallest, the system has the largest overall return, and each participant has the obvious possibility of cooperation. This provides a solution for multi-party participation in the energy supply market.
Abstract:
In view of the traditional manual feature method, which can not effectively extract the overall image deep information, a new method of scene classification based on deep learning feature fusion is proposed. Firstly, the grayscale symbiotic matrix (GLCM) and local two-value pattern (LBP) are used to extract the shallow information of texture features and local texture features with relevant spatial characteristics, and secondly, the deep information of images is extracted by the AlexNet migration learning network, and a 256-dimensional full-connect layer is added as feature output while the last layer of full connection layer is removed, and the two features are adaptively integrated The remote sensing images are classified and identified in the support vector machine (GS-SVM) optimized by the grid search algorithm. The experimental results of the 21 types of target data of the public dataset UC Merced and the 7 types of target data of RSSCN7 show that the average accuracy rate of the five experiments was 94.77% and93.79%, respectively. The experimental results are compared with other methods in the references and the classification accuracy under the same data conditions, which shows that the proposed method can effectively improve the classification accuracy of remote sensing image scenes.
In view of the traditional manual feature method, which can not effectively extract the overall image deep information, a new method of scene classification based on deep learning feature fusion is proposed. Firstly, the grayscale symbiotic matrix (GLCM) and local two-value pattern (LBP) are used to extract the shallow information of texture features and local texture features with relevant spatial characteristics, and secondly, the deep information of images is extracted by the AlexNet migration learning network, and a 256-dimensional full-connect layer is added as feature output while the last layer of full connection layer is removed, and the two features are adaptively integrated The remote sensing images are classified and identified in the support vector machine (GS-SVM) optimized by the grid search algorithm. The experimental results of the 21 types of target data of the public dataset UC Merced and the 7 types of target data of RSSCN7 show that the average accuracy rate of the five experiments was 94.77% and93.79%, respectively. The experimental results are compared with other methods in the references and the classification accuracy under the same data conditions, which shows that the proposed method can effectively improve the classification accuracy of remote sensing image scenes.
Abstract:
Previous studies on the fourth party logistics (4PL) routing problem have given little consideration to the risk preference of customer. Considering the customer"s risk preference, the value function in prospect theory is used to measure the customer"s risk attitude towards distribution cost and delivery period, and establish a mathematical model to maximize the utility of customer psychology evaluation. According to the characteristics of logistics network path optimization, an adaptive genetic algorithm embedded in Dijkstra algorithm is designed to solve the problem. Several numerical examples with different scales are designed, and the feasibility and effectiveness of the model and algorithm are illustrated on the basis of the simulation experiments. This paper provides a new decision-making model and optimization method for solving the fourth party logistics path optimization problem from the perspective of the influence of behavior on decision-making results.
Previous studies on the fourth party logistics (4PL) routing problem have given little consideration to the risk preference of customer. Considering the customer"s risk preference, the value function in prospect theory is used to measure the customer"s risk attitude towards distribution cost and delivery period, and establish a mathematical model to maximize the utility of customer psychology evaluation. According to the characteristics of logistics network path optimization, an adaptive genetic algorithm embedded in Dijkstra algorithm is designed to solve the problem. Several numerical examples with different scales are designed, and the feasibility and effectiveness of the model and algorithm are illustrated on the basis of the simulation experiments. This paper provides a new decision-making model and optimization method for solving the fourth party logistics path optimization problem from the perspective of the influence of behavior on decision-making results.
Abstract:
The output torque of a doubly salient electro-magnetic machine (DSEM) is related to the selection of the advance angle. Due to the nonlinear characteristics of DSEM, multiple finite element simulations and experiments are usually used to obtain the appropriate advance angle, and it needs to be reselected when the operating conditions change. To simplify the selection of the advance angle, an advance angle self-optimizing control of the DSEM drive system is proposed in this paper. Firstly, the relationship between the DSEM armature current enveloping area and the output torque is analyzed. With this for basis, the output torque estimation model is established. Then, the advance angle self-optimization control method is proposed. The advance angle of the DSEM drive system is perturbed by a fixed step. Based on the torque estimation model, by comparing the output torque of DSEM under different advance angles, the maximum output torque corresponding advance angle of the DSEM is automatically optimized. Finally, the effectiveness and feasibility of the proposed method are verified by field-circuit co-simulation and experiment on 12/8 DSEM.
The output torque of a doubly salient electro-magnetic machine (DSEM) is related to the selection of the advance angle. Due to the nonlinear characteristics of DSEM, multiple finite element simulations and experiments are usually used to obtain the appropriate advance angle, and it needs to be reselected when the operating conditions change. To simplify the selection of the advance angle, an advance angle self-optimizing control of the DSEM drive system is proposed in this paper. Firstly, the relationship between the DSEM armature current enveloping area and the output torque is analyzed. With this for basis, the output torque estimation model is established. Then, the advance angle self-optimization control method is proposed. The advance angle of the DSEM drive system is perturbed by a fixed step. Based on the torque estimation model, by comparing the output torque of DSEM under different advance angles, the maximum output torque corresponding advance angle of the DSEM is automatically optimized. Finally, the effectiveness and feasibility of the proposed method are verified by field-circuit co-simulation and experiment on 12/8 DSEM.
Abstract:
In the classification task of the fairly specialized chemical domain texts, it is hard to wholly characterize the spe-cialized phrases and some other phrases in the domain just by way of relying on the current word vector representation due to the specialized and complex diversity of the texts, which leads to the low accuracy of the classification task for chemical domain texts. To tackle these problems, the text classification model incorporating multi-granularity dy-namic semantic representations was proposed. Firstly, the adversarial perturbation was introduced into the word embedding layer of the model to enhance the ability of dynamic word vectors to represent the semantics. Then the word vector weights were redistributed by a multi-headed attention mechanism to obtain a better textual represen-tation of key semantic information. Finally, text representations of different granularities were extracted through the proposed multi-scale residual shrinkage deep pyramidal convolutional neural network (MSRS-DPCNN) and hybrid attention capsule bidirectional LSTM (HAC-BiLSTM) network model, fusing them and classifying the result of the fusion. Compared with existing models, the experimental results show that the proposed model achieves an F1-score of up to 84.62% on the chemical domain text dataset when using different word vector representations, an im-provement of 0.38 ~ 5.58 percentage points; The model also had better performance when it was evaluated for model generalization performance on the publicly available Chinese dataset THUCNews and the Tan Songbo hotel review dataset ChnSentiCorp.
In the classification task of the fairly specialized chemical domain texts, it is hard to wholly characterize the spe-cialized phrases and some other phrases in the domain just by way of relying on the current word vector representation due to the specialized and complex diversity of the texts, which leads to the low accuracy of the classification task for chemical domain texts. To tackle these problems, the text classification model incorporating multi-granularity dy-namic semantic representations was proposed. Firstly, the adversarial perturbation was introduced into the word embedding layer of the model to enhance the ability of dynamic word vectors to represent the semantics. Then the word vector weights were redistributed by a multi-headed attention mechanism to obtain a better textual represen-tation of key semantic information. Finally, text representations of different granularities were extracted through the proposed multi-scale residual shrinkage deep pyramidal convolutional neural network (MSRS-DPCNN) and hybrid attention capsule bidirectional LSTM (HAC-BiLSTM) network model, fusing them and classifying the result of the fusion. Compared with existing models, the experimental results show that the proposed model achieves an F1-score of up to 84.62% on the chemical domain text dataset when using different word vector representations, an im-provement of 0.38 ~ 5.58 percentage points; The model also had better performance when it was evaluated for model generalization performance on the publicly available Chinese dataset THUCNews and the Tan Songbo hotel review dataset ChnSentiCorp.
Abstract:
Scientific evaluation of urbanization quality is of great significance to measure the level of urban development and promote urban sustainable development. Based on the two dimensions of urban development and urban environmental quality, the evaluation index system of urbanization quality in Liaoning Province was constructed, and the entropy weight method, standard deviation ellipse and geographic detector were comprehensively used to explore the urbanization quality and spatial distribution law, spatial evolution trend and influence factor detection in Liaoning Province. The results show that: (1) From 2005 to 2019, the overall urbanization quality index increased from 2.74 to 4.74, the quality of economic urbanization subsystem has the largest increase, the quality of social urbanization and environmental governance subsystem has an obvious upward trend, and the quality of environmental state and environmental pressure subsystem is relatively stable. (2) The urbanization quality index of each city ranges from 0.087 to 0.740, and the spatial distribution shows that Shenyang is the "big core" and Dalian is the "small core". (3) The center of gravity ellipse of urbanization quality standard deviation in the study area is located in Liaoyang City, and the deflection trajectory is "southeast, northwest, southwest" from 2005 to 2019, and the distance and speed of the center of gravity ellipse of standard deviation increase year by year. (4) Subsystems are mainly affecting the quality of urbanization in Liaoning province economic urbanization, social urbanization and environmental control factors, specific to the index layer, affect the comprehensive index of urbanization area is main, the whole society fixed assets investment, to attend primary endowment insurance number, the number of college students per ten thousand people, the Domestic garbage volume and urban daily sewage treatment capacity, And their influence continues to grow.
Scientific evaluation of urbanization quality is of great significance to measure the level of urban development and promote urban sustainable development. Based on the two dimensions of urban development and urban environmental quality, the evaluation index system of urbanization quality in Liaoning Province was constructed, and the entropy weight method, standard deviation ellipse and geographic detector were comprehensively used to explore the urbanization quality and spatial distribution law, spatial evolution trend and influence factor detection in Liaoning Province. The results show that: (1) From 2005 to 2019, the overall urbanization quality index increased from 2.74 to 4.74, the quality of economic urbanization subsystem has the largest increase, the quality of social urbanization and environmental governance subsystem has an obvious upward trend, and the quality of environmental state and environmental pressure subsystem is relatively stable. (2) The urbanization quality index of each city ranges from 0.087 to 0.740, and the spatial distribution shows that Shenyang is the "big core" and Dalian is the "small core". (3) The center of gravity ellipse of urbanization quality standard deviation in the study area is located in Liaoyang City, and the deflection trajectory is "southeast, northwest, southwest" from 2005 to 2019, and the distance and speed of the center of gravity ellipse of standard deviation increase year by year. (4) Subsystems are mainly affecting the quality of urbanization in Liaoning province economic urbanization, social urbanization and environmental control factors, specific to the index layer, affect the comprehensive index of urbanization area is main, the whole society fixed assets investment, to attend primary endowment insurance number, the number of college students per ten thousand people, the Domestic garbage volume and urban daily sewage treatment capacity, And their influence continues to grow.
Abstract:
The sparse and low mangroves are difficult to be accurately extracted due to the periodic changes of the tide level. Taking the Beibu Gulf of Guangxi as the research area, the paper constructed a mangrove identification decision tree model using Landsat8 OLI images at low and high tidal levels and DEM (Digital Elevation Model) data, and took SVM (support vector machine) method to evaluate the feasibility of the decision tree method. The research results show that: (1) The spectra of mangroves were different due to the height and density of mangroves canopy and the tidal levels. There was a serious "same spectrum with different species" effect between sparse and low-lying mangroves and the water-terrestrial vegetation mixed pixel and shady slope forest. (2) Mangroves were classified into high-density mangroves and sparse-dwarf mangroves according to the differences in height and density of canopy. Whether using low-tide, high-tide or multi-tide images for the SVM methods, this can bring about that the overall accuracy values (Mangroves were divided into mangrove and non-mangrove categories) were improved by 4.65%, 4.41% and 7.22% respectively, (3) The overall accuracy and Kappa coefficient of decision tree model based on multi-tide image and DEM were 98.80% and 0.973 respectively, which were 1.62% and 0.035 higher than the best value in SVM method. Therefore, considering the characteristics of mangrove height, density, tide level and DEM could significantly improve the accuracy of remote sensing identification of mangroves.
The sparse and low mangroves are difficult to be accurately extracted due to the periodic changes of the tide level. Taking the Beibu Gulf of Guangxi as the research area, the paper constructed a mangrove identification decision tree model using Landsat8 OLI images at low and high tidal levels and DEM (Digital Elevation Model) data, and took SVM (support vector machine) method to evaluate the feasibility of the decision tree method. The research results show that: (1) The spectra of mangroves were different due to the height and density of mangroves canopy and the tidal levels. There was a serious "same spectrum with different species" effect between sparse and low-lying mangroves and the water-terrestrial vegetation mixed pixel and shady slope forest. (2) Mangroves were classified into high-density mangroves and sparse-dwarf mangroves according to the differences in height and density of canopy. Whether using low-tide, high-tide or multi-tide images for the SVM methods, this can bring about that the overall accuracy values (Mangroves were divided into mangrove and non-mangrove categories) were improved by 4.65%, 4.41% and 7.22% respectively, (3) The overall accuracy and Kappa coefficient of decision tree model based on multi-tide image and DEM were 98.80% and 0.973 respectively, which were 1.62% and 0.035 higher than the best value in SVM method. Therefore, considering the characteristics of mangrove height, density, tide level and DEM could significantly improve the accuracy of remote sensing identification of mangroves.
Abstract:
Gaseous ammonia is one of the main precursors for generating PM2.5 in the atmosphere. In order to identifying the sources of ammonia more accurately with isotopic technique,and to alleviating PM2.5 pollution in urban air, relevant data published in recent years have been collected and analyzed on sampling and quantifying of gaseous ammonia, determination or estimating of its nitrogen isotope ratio value, the nitrogen isotope compositions and fractionations for ammonia both in ambient air and emitted from different sources, the isotopic compositions of ammonium in atmospheric particulate matter and rainfall, as well as source apportionment of atmospheric ammonia. Several suggestions for future research have been accordingly proposed：The least amount of ammonia should be determined before collecting air samples with passive method in case of nitrogen isotope fractionation. Study may be strengthened on the nitrogen isotope compositions of ammonia from biomass combustion, natural soil, ocean, sewage plant, vegetation, and other potential sources. The mechanism on isotopic variation for ammonia from same emitting source could be further investigated. Both concentrations and isotopic compositions of gaseous ammonia and particulate ammonium perhaps may be simultaneously measured at higher temporal resolution,to understand the mechanism for nitrogen isotope fractionation in the process of ammonium forming. The isotopic fractionation of gaseous ammonia also should be deeply researched under different meteorological conditions and/or air contamination status.
Gaseous ammonia is one of the main precursors for generating PM2.5 in the atmosphere. In order to identifying the sources of ammonia more accurately with isotopic technique,and to alleviating PM2.5 pollution in urban air, relevant data published in recent years have been collected and analyzed on sampling and quantifying of gaseous ammonia, determination or estimating of its nitrogen isotope ratio value, the nitrogen isotope compositions and fractionations for ammonia both in ambient air and emitted from different sources, the isotopic compositions of ammonium in atmospheric particulate matter and rainfall, as well as source apportionment of atmospheric ammonia. Several suggestions for future research have been accordingly proposed：The least amount of ammonia should be determined before collecting air samples with passive method in case of nitrogen isotope fractionation. Study may be strengthened on the nitrogen isotope compositions of ammonia from biomass combustion, natural soil, ocean, sewage plant, vegetation, and other potential sources. The mechanism on isotopic variation for ammonia from same emitting source could be further investigated. Both concentrations and isotopic compositions of gaseous ammonia and particulate ammonium perhaps may be simultaneously measured at higher temporal resolution,to understand the mechanism for nitrogen isotope fractionation in the process of ammonium forming. The isotopic fractionation of gaseous ammonia also should be deeply researched under different meteorological conditions and/or air contamination status.
Abstract:
With the rapid development of ICV(Intelligent Connected Vehicle) industry, data exchange between vehicle and human, vehicle and vehicle as well as between vehicle and external environment has become nor mal, automobile safety has been seriously threatened, and it is very important to study automobile safety certification. Safety certification of vehicle-mounted PEPS (Passive Entry Passive Start) and EMS (Engine Management System) determines the safety performance of the vehicle, which is the prerequisite to ensure the safe operation of the vehicle. At present, 128bits AES algorithm is widely used in the industry to realize the security certification of PEPS and EMS. Because the AES key generation and key scheduling algorithm is more complex than the national encryption algorithm SM4, the encryption and decryption time is longer, and the amount of code implemented by the algorithm is larger, which takes up too much MCU resources. Therefore, it is proposed to apply the national secret SM4 algorithm to the safety certification of on-board PEPS and EMS, shorten the encryption and decryption time, and effectively improve the efficiency of data transmission. At the same time, the high level language is used to implement the algorithm and transplanted to the domestic MCU GD32F103 to realize localization of the product and reduce the cost. The state secret algorithm SM4 is extended to provide a research basis for ICV security certification.
With the rapid development of ICV(Intelligent Connected Vehicle) industry, data exchange between vehicle and human, vehicle and vehicle as well as between vehicle and external environment has become nor mal, automobile safety has been seriously threatened, and it is very important to study automobile safety certification. Safety certification of vehicle-mounted PEPS (Passive Entry Passive Start) and EMS (Engine Management System) determines the safety performance of the vehicle, which is the prerequisite to ensure the safe operation of the vehicle. At present, 128bits AES algorithm is widely used in the industry to realize the security certification of PEPS and EMS. Because the AES key generation and key scheduling algorithm is more complex than the national encryption algorithm SM4, the encryption and decryption time is longer, and the amount of code implemented by the algorithm is larger, which takes up too much MCU resources. Therefore, it is proposed to apply the national secret SM4 algorithm to the safety certification of on-board PEPS and EMS, shorten the encryption and decryption time, and effectively improve the efficiency of data transmission. At the same time, the high level language is used to implement the algorithm and transplanted to the domestic MCU GD32F103 to realize localization of the product and reduce the cost. The state secret algorithm SM4 is extended to provide a research basis for ICV security certification.
Abstract:
Video coding techniques effectively address the data volume problem of raw video, however, the achieved compression efficiency comes at the cost of video quality degradation. To improve the visual quality of compressed video, a Detail Recovery Convolutional Neural Network (DRCNN)-based video quality enhancement method is proposed in this paper, which consists of a main denoising branch and a detail compensation branch. To effectively extract and eliminate the compression distortions, a main denoising branch is proposed, in which a Multi-Scale Distortion Feature Extraction Block (MDFEBs) is proposed to pay more attention to the distortion areas in the compressed video, and improve the distortion feature learning ability of the proposed DRCNN. Furthermore, to enrich the details in the compressed video, a detail compensation branch is proposed, which firstly adopts a content feature extractor composed of a pre-trained ResNet-50 to provide abundant content features, such as salient objects, shapes, details and so on, and then a Detail Response Block (DRB) is designed to efficiently extract the detailed features from the content features. Extensive experimental results show that the proposed DRCNN achieves the best compressed video quality enhancement performance as compared with the state-of-the-art methods.
Video coding techniques effectively address the data volume problem of raw video, however, the achieved compression efficiency comes at the cost of video quality degradation. To improve the visual quality of compressed video, a Detail Recovery Convolutional Neural Network (DRCNN)-based video quality enhancement method is proposed in this paper, which consists of a main denoising branch and a detail compensation branch. To effectively extract and eliminate the compression distortions, a main denoising branch is proposed, in which a Multi-Scale Distortion Feature Extraction Block (MDFEBs) is proposed to pay more attention to the distortion areas in the compressed video, and improve the distortion feature learning ability of the proposed DRCNN. Furthermore, to enrich the details in the compressed video, a detail compensation branch is proposed, which firstly adopts a content feature extractor composed of a pre-trained ResNet-50 to provide abundant content features, such as salient objects, shapes, details and so on, and then a Detail Response Block (DRB) is designed to efficiently extract the detailed features from the content features. Extensive experimental results show that the proposed DRCNN achieves the best compressed video quality enhancement performance as compared with the state-of-the-art methods.
Abstract:
Traditional AEC’s performance is restricted when using double-talk detector to determine the double- talk and single-talk scenarios. Blind source separation signal model is a full duplex model with both far-end and near-end signals, so AEC based on BSS does not need the double-talk detector. This paper adopts auxiliary function based independent component analysis algorithm to realize acoustic echo cancellation in frequency domain. The object function of Aux-ICA is minimizing the mutual information, and the auxiliary function technique is used for the optimization. Simulation results show that this method has lower computational complexity and better acoustic echo cancellation performance in the continuous double-talk scenarios.
Traditional AEC’s performance is restricted when using double-talk detector to determine the double- talk and single-talk scenarios. Blind source separation signal model is a full duplex model with both far-end and near-end signals, so AEC based on BSS does not need the double-talk detector. This paper adopts auxiliary function based independent component analysis algorithm to realize acoustic echo cancellation in frequency domain. The object function of Aux-ICA is minimizing the mutual information, and the auxiliary function technique is used for the optimization. Simulation results show that this method has lower computational complexity and better acoustic echo cancellation performance in the continuous double-talk scenarios.
Abstract:
Since the outbreak of COVID-19, the impact of small and medium-sized service enterprises has been particularly strong. Based on the background that local government and small and medium-sized service enterprises jointly issue consumption vouchers to deal with the epidemic situation, this paper constructs a tripartite evolutionary game model of local government, small and medium-sized service enterprises and consumers, and studies the influence factors of government paid subsidies, credibility, consumption voucher deduction amount and other factors on the strategies of each stakeholder, in order to realize the benign cooperation between local governments and small and medium-sized service enterprises, stimulate consumption and benefit the people. The results show that under the government paid support strategy, the subsidy amount is low, and the paid support strategy has a greater impact on the credibility, which will prompt the government to choose paid support for enterprises; Under the positive self-help strategy, the enterprises are more willing to adhere to the positive self-help strategy with higher paid subsidies and less deduction for consumption vouchers; The impact of consumption vouchers on the actual consumption amount of consumers is small, and the positive self-help enterprises have more deductions for consumption vouchers, which will encourage consumers to choose voucher consumption; The implementation of differentiated support policies, combined with the accurate issuance of consumption vouchers, can promote economic recovery.
Since the outbreak of COVID-19, the impact of small and medium-sized service enterprises has been particularly strong. Based on the background that local government and small and medium-sized service enterprises jointly issue consumption vouchers to deal with the epidemic situation, this paper constructs a tripartite evolutionary game model of local government, small and medium-sized service enterprises and consumers, and studies the influence factors of government paid subsidies, credibility, consumption voucher deduction amount and other factors on the strategies of each stakeholder, in order to realize the benign cooperation between local governments and small and medium-sized service enterprises, stimulate consumption and benefit the people. The results show that under the government paid support strategy, the subsidy amount is low, and the paid support strategy has a greater impact on the credibility, which will prompt the government to choose paid support for enterprises; Under the positive self-help strategy, the enterprises are more willing to adhere to the positive self-help strategy with higher paid subsidies and less deduction for consumption vouchers; The impact of consumption vouchers on the actual consumption amount of consumers is small, and the positive self-help enterprises have more deductions for consumption vouchers, which will encourage consumers to choose voucher consumption; The implementation of differentiated support policies, combined with the accurate issuance of consumption vouchers, can promote economic recovery.
Abstract:
Aiming at the collaborative optimization problem of multi-UAV reconnaissance and communication services for multiple heterogeneous targets, by considering the mission requirements and value of different targets, as well as the multi-UAV coordination gain and behavior restriction relationship, the Stackelberg game model is constructed. The upper-level drone is established as the leader of the game, the lower-level drone is established as the follower of the game, and a distributed strategy update iterative algorithm is proposed, which realizes the stable convergence of the multi-UAV task allocation scheme and the optimization of the task revenue. The simulation results show that the proposed method can effectively improve the efficiency of multi-UAV systems to complete multiple tasks at the same time, and can achieve efficient collaboration for the value of heterogeneous tasks in different environments.
Aiming at the collaborative optimization problem of multi-UAV reconnaissance and communication services for multiple heterogeneous targets, by considering the mission requirements and value of different targets, as well as the multi-UAV coordination gain and behavior restriction relationship, the Stackelberg game model is constructed. The upper-level drone is established as the leader of the game, the lower-level drone is established as the follower of the game, and a distributed strategy update iterative algorithm is proposed, which realizes the stable convergence of the multi-UAV task allocation scheme and the optimization of the task revenue. The simulation results show that the proposed method can effectively improve the efficiency of multi-UAV systems to complete multiple tasks at the same time, and can achieve efficient collaboration for the value of heterogeneous tasks in different environments.
Abstract:
The development of 5G network promotes the strategic transformation and upgrading of the vertical industry of the industrial Internet. The 5G SA (Stand-Alone) network sinks the user plane function (UPF) into the business area, and the control plane remains in the center of the large area. All business data is accessed through the 5G edge computing interface (MEC) to the business application server, and all traffic for users to access their own servers is completed internally, which meets network security requirements and minimizes business path delay. Through optimized particle swarm optimization (PSO), a fitness function based on network physical node resources and link resources is constructed, and the Bayesian evaluation method is introduced to calculate the physical node isolation factor. A threshold is set as the average isolation factor of all nodes as a constraint condition to ensure the isolation performance of network slices, and iterative solution is used to adjust the path resources between network slices to improve network performance. After the actual 5G network test of the Intelligence mine, the results show that the algorithm can increase the revenue-cost ratio by 10%, increase the link utilization rate by 19%, and increase the uplink rate by 20%-50%, which fully guarantees the service quality of the mining area network.
The development of 5G network promotes the strategic transformation and upgrading of the vertical industry of the industrial Internet. The 5G SA (Stand-Alone) network sinks the user plane function (UPF) into the business area, and the control plane remains in the center of the large area. All business data is accessed through the 5G edge computing interface (MEC) to the business application server, and all traffic for users to access their own servers is completed internally, which meets network security requirements and minimizes business path delay. Through optimized particle swarm optimization (PSO), a fitness function based on network physical node resources and link resources is constructed, and the Bayesian evaluation method is introduced to calculate the physical node isolation factor. A threshold is set as the average isolation factor of all nodes as a constraint condition to ensure the isolation performance of network slices, and iterative solution is used to adjust the path resources between network slices to improve network performance. After the actual 5G network test of the Intelligence mine, the results show that the algorithm can increase the revenue-cost ratio by 10%, increase the link utilization rate by 19%, and increase the uplink rate by 20%-50%, which fully guarantees the service quality of the mining area network.
Abstract:
Nitrogen-doped Co3O4 nanosheets (N-Co NS) were synthesized by sacrificial template method. The morphological structure and chemical composition of the obtained materials were characterized by Transmission electron microscopy (TEM), Atomic force microscopy (AFM) and X-ray photoelectron spectroscopy (XPS). In addition, the catalytic performance of the prepared catalysts was evaluated by catalytically activating peroxymonosulfate (PMS) to degrade bisphenol A (BPA) in water. Compared with Co3O4 nanoparticles (Co NP) and Co3O4 nanosheets (Co NS), N-Co NS exhibits higher catalytic performance according to the experimental results. Under the reaction conditions that the dosage of PMS is 2 mM and the initial concentration of BPA is 50 mg L-1, N-Co NS completely degrades BPA in water within 10 minutes, indicating that N-doping and two-dimensional nanosheet structure are beneficial to the improvement of catalyst performance. N-Co NS still has high activity in complex water chemical environment proved by the pH and ions effect experiments. Besides, the high oxidative activity hydroxyl radicals and sulfate radicals were produced in the reaction system, which was confirmed by the trapping experiments and Electron paramagnetic resonance (EPR) tests.
Nitrogen-doped Co3O4 nanosheets (N-Co NS) were synthesized by sacrificial template method. The morphological structure and chemical composition of the obtained materials were characterized by Transmission electron microscopy (TEM), Atomic force microscopy (AFM) and X-ray photoelectron spectroscopy (XPS). In addition, the catalytic performance of the prepared catalysts was evaluated by catalytically activating peroxymonosulfate (PMS) to degrade bisphenol A (BPA) in water. Compared with Co3O4 nanoparticles (Co NP) and Co3O4 nanosheets (Co NS), N-Co NS exhibits higher catalytic performance according to the experimental results. Under the reaction conditions that the dosage of PMS is 2 mM and the initial concentration of BPA is 50 mg L-1, N-Co NS completely degrades BPA in water within 10 minutes, indicating that N-doping and two-dimensional nanosheet structure are beneficial to the improvement of catalyst performance. N-Co NS still has high activity in complex water chemical environment proved by the pH and ions effect experiments. Besides, the high oxidative activity hydroxyl radicals and sulfate radicals were produced in the reaction system, which was confirmed by the trapping experiments and Electron paramagnetic resonance (EPR) tests.
Abstract:
In view of the problems that aspect level sentiment analysis tasks can not give full consideration to syntactic comprehensiveness and semantic relevance, and the graph volume used in most studies only considers the top-down dissemination of information and ignores the bottom-up aggregation of information, this paper proposes an sentiment analysis model based on attention and dual channel network. While expanding the dependency representation, the model uses self attention to obtain the information matrix with semantic relevance, and uses a dual channel network to combine comprehensive syntactic and semantic relevance information. The dual channel network focuses on the semantic features of top-down propagation and the structural features of bottom-up aggregation respectively. The graph convolution output in the channel will interact with the information matrix, pay attention to complement the residual, and then complete the tasks in the channel through average pooling. Finally, the final emotion classification features are obtained by the fusion of semantic based and structure based decision-making. The experimental results show that the accuracy and F1 value of the model are improved on three public data sets.
In view of the problems that aspect level sentiment analysis tasks can not give full consideration to syntactic comprehensiveness and semantic relevance, and the graph volume used in most studies only considers the top-down dissemination of information and ignores the bottom-up aggregation of information, this paper proposes an sentiment analysis model based on attention and dual channel network. While expanding the dependency representation, the model uses self attention to obtain the information matrix with semantic relevance, and uses a dual channel network to combine comprehensive syntactic and semantic relevance information. The dual channel network focuses on the semantic features of top-down propagation and the structural features of bottom-up aggregation respectively. The graph convolution output in the channel will interact with the information matrix, pay attention to complement the residual, and then complete the tasks in the channel through average pooling. Finally, the final emotion classification features are obtained by the fusion of semantic based and structure based decision-making. The experimental results show that the accuracy and F1 value of the model are improved on three public data sets.
Abstract:
In order to explore the impact of climate change on the lake area of Silingco, this paper based on Landsat satellite remote sensing image data（1988-2020）, the information on the lake area of Silingco was obtained by using the method of maximum likelihood method, then the change characteristics of the lake area and climate were analyzed according to the information on the lake area and basin meteorological data using Mann-Kendall tests and linear regression. The correlation between the lake area and the meteorological factors was also analyzed using Pearson correlation. The results showed that, in the past 30 years, the lake area of Silingco increased by 650.70 km2 at the rate of 255.10km2/10a. The northern part and southern part of the lake changed obviously. The mean annual temperature and precipitation increased significantly during 1985-2020 (p < 0.05) and the mean maximum snow depth decreased significantly (p < 0.05)，and the change rate was 0.50°C/10a for temperature,17.32mm/10a for precipitation and 0.65cm/10a for the maximum depth of snowcover, the changes of lake area increased significantly with both the rise of air temperature in the basin and decreasing maximum depth of snow cover in the cold season(p<0.001), therefore, the main reason for the lake's water area expansion of Silingco was the rise of the air temperature which led to the increase of water supply to Silingco mainlyfrom snow and ice meltwater.
In order to explore the impact of climate change on the lake area of Silingco, this paper based on Landsat satellite remote sensing image data（1988-2020）, the information on the lake area of Silingco was obtained by using the method of maximum likelihood method, then the change characteristics of the lake area and climate were analyzed according to the information on the lake area and basin meteorological data using Mann-Kendall tests and linear regression. The correlation between the lake area and the meteorological factors was also analyzed using Pearson correlation. The results showed that, in the past 30 years, the lake area of Silingco increased by 650.70 km2 at the rate of 255.10km2/10a. The northern part and southern part of the lake changed obviously. The mean annual temperature and precipitation increased significantly during 1985-2020 (p < 0.05) and the mean maximum snow depth decreased significantly (p < 0.05)，and the change rate was 0.50°C/10a for temperature,17.32mm/10a for precipitation and 0.65cm/10a for the maximum depth of snowcover, the changes of lake area increased significantly with both the rise of air temperature in the basin and decreasing maximum depth of snow cover in the cold season(p<0.001), therefore, the main reason for the lake's water area expansion of Silingco was the rise of the air temperature which led to the increase of water supply to Silingco mainlyfrom snow and ice meltwater.
Abstract:
In order to solve the problem of simultaneous multi-site PM2.5 prediction, a hierarchical autoregressive spatio-temporal model is proposed under the Bayesian framework. The true daily average concentration of PM2.5 is regarded as a potential spatio-temporal process. The first-order autoregressive process is used to describe the temporal correlation, and the spatial correlation is captured based on the Matérn process, which greatly improves the efficiency of dimension reduction and synchronous prediction. In addition, meteorological factors such as daily maximum temperature, relative humidity and wind speed are used as explanatory variables to improve the accuracy of prediction. Thanks to the hierarchical structure of the model, the parameter estimation and prediction process can be realized by Bayesian method and MCMC. The empirical analysis of daily PM2.5 concentration in Beijing shows that the proposed model has good interpolation or prediction effect in spatial and temporal dimensions.
In order to solve the problem of simultaneous multi-site PM2.5 prediction, a hierarchical autoregressive spatio-temporal model is proposed under the Bayesian framework. The true daily average concentration of PM2.5 is regarded as a potential spatio-temporal process. The first-order autoregressive process is used to describe the temporal correlation, and the spatial correlation is captured based on the Matérn process, which greatly improves the efficiency of dimension reduction and synchronous prediction. In addition, meteorological factors such as daily maximum temperature, relative humidity and wind speed are used as explanatory variables to improve the accuracy of prediction. Thanks to the hierarchical structure of the model, the parameter estimation and prediction process can be realized by Bayesian method and MCMC. The empirical analysis of daily PM2.5 concentration in Beijing shows that the proposed model has good interpolation or prediction effect in spatial and temporal dimensions.
Abstract:
Increasingly frequent bird activities have brought serious threat on the safe operation of transmission lines, and the existing audio bird-repelling device can not perennially effectively drive birds due to the lack of intellectuality. In order to solve the above problems, this paper presents an audio bird-repelling strategy based on improved Q-learning algorithm. First of all, in order to evaluate the effect of each audio, the behavior of birds after hearing the audio was quantified into different bird response types by combining with the fuzzy theory. Then, an audio bird-repelling experiment was designed, the data of each audio bird-repelling effect is counted, and the initial weight of each audio is obtained, which provided experimental basis for the audio selection of audio bird-repelling device. In order to make the audio weight more consistent with the actual experimental situation, the weight calculation formula of CRITIC(Criteria Importance Though Intercrieria Correlation) is optimized. Finally, the q-learning algorithm is improved by using the audio weight which is obtained from the experiment, and the contrast experiment with other audio bird-repelling strategies is designed. Experimental data show that the effect of the improved Q-learning algorithm for audio bird-repelling strategy is better than the other audio bird-repelling strategies, which has fast convergence, stable bird-repelling effect, and can reduce the adaptability of birds.
Increasingly frequent bird activities have brought serious threat on the safe operation of transmission lines, and the existing audio bird-repelling device can not perennially effectively drive birds due to the lack of intellectuality. In order to solve the above problems, this paper presents an audio bird-repelling strategy based on improved Q-learning algorithm. First of all, in order to evaluate the effect of each audio, the behavior of birds after hearing the audio was quantified into different bird response types by combining with the fuzzy theory. Then, an audio bird-repelling experiment was designed, the data of each audio bird-repelling effect is counted, and the initial weight of each audio is obtained, which provided experimental basis for the audio selection of audio bird-repelling device. In order to make the audio weight more consistent with the actual experimental situation, the weight calculation formula of CRITIC(Criteria Importance Though Intercrieria Correlation) is optimized. Finally, the q-learning algorithm is improved by using the audio weight which is obtained from the experiment, and the contrast experiment with other audio bird-repelling strategies is designed. Experimental data show that the effect of the improved Q-learning algorithm for audio bird-repelling strategy is better than the other audio bird-repelling strategies, which has fast convergence, stable bird-repelling effect, and can reduce the adaptability of birds.
Abstract:
The operation experience shows that the root cause of the harm of large-scale DC magnetic bias caused by DC grounding electrode lies in the uneven surface potential distribution. The Chebyshev polynomial is used to fit the Hankel transform kernel function of surface potential in order to solve the problem of solving the surface potential distribution in the complex earth model of wide area depth stratification. Through the shift operation, coefficient expansion and truncation error determination of Chebyshev polynomial, the adaptive order fitting method of Chebyshev polynomial for kernel function is obtained, which greatly reduces the calculation difficulty of surface potential distribution in a large area caused by DC grounding electrode. Compared with the standard grounding calculation software CDEGS, the earth surface potential deviation is less than 1 V in the range of 1 km ~ 100 km when the DC grounding current is 5 000 A. The influence of the Chebyshev polynomial series on the results is further analyzed, and it is confirmed that the accuracy of the 20-order Chebyshev polynomial can meet the application requirements of the general DC bias risk assessment. The surface potential rapid assessment method based on the shifted Chebyshev polynomial provides a basic technical means for the risk assessment of DC bias, which is helpful to reduce the difficulty of risk assessment of DC bias in power grid.
The operation experience shows that the root cause of the harm of large-scale DC magnetic bias caused by DC grounding electrode lies in the uneven surface potential distribution. The Chebyshev polynomial is used to fit the Hankel transform kernel function of surface potential in order to solve the problem of solving the surface potential distribution in the complex earth model of wide area depth stratification. Through the shift operation, coefficient expansion and truncation error determination of Chebyshev polynomial, the adaptive order fitting method of Chebyshev polynomial for kernel function is obtained, which greatly reduces the calculation difficulty of surface potential distribution in a large area caused by DC grounding electrode. Compared with the standard grounding calculation software CDEGS, the earth surface potential deviation is less than 1 V in the range of 1 km ~ 100 km when the DC grounding current is 5 000 A. The influence of the Chebyshev polynomial series on the results is further analyzed, and it is confirmed that the accuracy of the 20-order Chebyshev polynomial can meet the application requirements of the general DC bias risk assessment. The surface potential rapid assessment method based on the shifted Chebyshev polynomial provides a basic technical means for the risk assessment of DC bias, which is helpful to reduce the difficulty of risk assessment of DC bias in power grid.
Abstract:
Aiming at the current problems of low utilization efficiency and low income caused by high energy storage cost and lack of trading platform, a non-cooperative game trading model based on blockchain technology is designed. The concept of shared energy storage is introduced by taking microgrids as nodes, and the nodes identity verification is carried out by using digital signature technology to effectively improve the security of transaction. Based on the self-driven and self-executing characteristics of smart contract, a shared energy storage trading mechanism is established. Through the establishment of non-cooperative game model, the benefits of each node can be maximized, energy storage benefits can be improved, the enthusiasm of node users to participate in transactions can be improved, and the industrial structure can be further optimized. Example analysis shows that the non-cooperative game trading model of shared energy storage based on blockchain proposed in this paper can effectively realize energy storage participation in market transactions, improve energy storage utilization efficiency, increase energy storage project income sources, and provide effective technical and theoretical support for the further development of energy storage industry.
Aiming at the current problems of low utilization efficiency and low income caused by high energy storage cost and lack of trading platform, a non-cooperative game trading model based on blockchain technology is designed. The concept of shared energy storage is introduced by taking microgrids as nodes, and the nodes identity verification is carried out by using digital signature technology to effectively improve the security of transaction. Based on the self-driven and self-executing characteristics of smart contract, a shared energy storage trading mechanism is established. Through the establishment of non-cooperative game model, the benefits of each node can be maximized, energy storage benefits can be improved, the enthusiasm of node users to participate in transactions can be improved, and the industrial structure can be further optimized. Example analysis shows that the non-cooperative game trading model of shared energy storage based on blockchain proposed in this paper can effectively realize energy storage participation in market transactions, improve energy storage utilization efficiency, increase energy storage project income sources, and provide effective technical and theoretical support for the further development of energy storage industry.
Abstract:
Aiming at the optimization problem of multi-agent systems, a distributed optimization algorithm based on dynamic event triggered mechanism is proposed in this paper. A new dynamic event triggered controller is designed based on Lyapunov function method. Compared with the traditional static triggered control methods, the proposed algorithm can effectively reduce the communication burden between agents and the calculation burden of controllers. In addition, as the periodic sampling information is used to design the event triggered condition, it is unrequired to continuously detect the event triggered condition. Moreover, Zeno behavior can be avoided. The effectiveness of the algorithm is verified by a numerical simulation.
Aiming at the optimization problem of multi-agent systems, a distributed optimization algorithm based on dynamic event triggered mechanism is proposed in this paper. A new dynamic event triggered controller is designed based on Lyapunov function method. Compared with the traditional static triggered control methods, the proposed algorithm can effectively reduce the communication burden between agents and the calculation burden of controllers. In addition, as the periodic sampling information is used to design the event triggered condition, it is unrequired to continuously detect the event triggered condition. Moreover, Zeno behavior can be avoided. The effectiveness of the algorithm is verified by a numerical simulation.
Abstract:
This paper proposed a Marching Trapezoidal Polyhedrons 3D modeling algorithm (MTPD) based on the spatial cone-shape distribution of Doppler weather radar base data. In this algorithm, a trapezoidal polyhedron was introduced as a basic volume element for modelling. On the other hand, the hexahedral ?index or tetrahedral index was selected as the construction model for the 3D iso-surface based on the difference in spatial range for 3D radar modeling, to balance the efficiency of the algorithm and the precision of the modeling results. Based on this algorithm, a doppler weather radar 3D visualization platform was developed using WebGL technology. The results revealed that the algorithm significantly improved efficiency without compromising the precision of 3D modeling when compared with the conventional modeling algorithm based on radar grid data. The computational efficiency for the algorithms were increased by 62.5% and 23.0%, under the tetrahedral index mode, the 3D echo structure was more continuous, with a higher level of precision. The doppler weather 3D radar visualization platform based on the B/S architecture could provide a cross-platform 3D radar display, thus visualizing the 3D structure of convective cloud effectively.
This paper proposed a Marching Trapezoidal Polyhedrons 3D modeling algorithm (MTPD) based on the spatial cone-shape distribution of Doppler weather radar base data. In this algorithm, a trapezoidal polyhedron was introduced as a basic volume element for modelling. On the other hand, the hexahedral ?index or tetrahedral index was selected as the construction model for the 3D iso-surface based on the difference in spatial range for 3D radar modeling, to balance the efficiency of the algorithm and the precision of the modeling results. Based on this algorithm, a doppler weather radar 3D visualization platform was developed using WebGL technology. The results revealed that the algorithm significantly improved efficiency without compromising the precision of 3D modeling when compared with the conventional modeling algorithm based on radar grid data. The computational efficiency for the algorithms were increased by 62.5% and 23.0%, under the tetrahedral index mode, the 3D echo structure was more continuous, with a higher level of precision. The doppler weather 3D radar visualization platform based on the B/S architecture could provide a cross-platform 3D radar display, thus visualizing the 3D structure of convective cloud effectively.
Abstract:
In order to explore the impact of different plant communities on atmospheric particulate matter concentration,taking Jinshui District of Zhengzhou City as an example, in the winter of 2020 (December 2020-February 2021), there will The PM2.5 and PM10 concentrations and meteorological factors (temperature, humidity and wind speed) of the plant communities in Lanbaowan) and the Cultural and Educational District (Henan Agricultural University) were monitored. The results indicated that the diurnal variation trend of PM2.5 and PM10 concentrations was basically the same between different places in each functional area.generally showing early high and late low; there are significant differences in PM2.5 and PM10 concentrations between different plant community structures, and the difference between the square plot and other plots is the most significant; In the three functional areas, the blocking rates of PM2.5 and PM10 were all the highest in arbor, shrub and grass structure, followed by arbor and grass structure, which were mostly higher in arbor and grass plot than in arbor and grass plot, and the lowest in shrub and grass plot and lawn; PM2.5, PM10 concentration has a negative correlation with temperature, a positive correlation with relative humidity, and a negative correlation with wind speed.
In order to explore the impact of different plant communities on atmospheric particulate matter concentration,taking Jinshui District of Zhengzhou City as an example, in the winter of 2020 (December 2020-February 2021), there will The PM2.5 and PM10 concentrations and meteorological factors (temperature, humidity and wind speed) of the plant communities in Lanbaowan) and the Cultural and Educational District (Henan Agricultural University) were monitored. The results indicated that the diurnal variation trend of PM2.5 and PM10 concentrations was basically the same between different places in each functional area.generally showing early high and late low; there are significant differences in PM2.5 and PM10 concentrations between different plant community structures, and the difference between the square plot and other plots is the most significant; In the three functional areas, the blocking rates of PM2.5 and PM10 were all the highest in arbor, shrub and grass structure, followed by arbor and grass structure, which were mostly higher in arbor and grass plot than in arbor and grass plot, and the lowest in shrub and grass plot and lawn; PM2.5, PM10 concentration has a negative correlation with temperature, a positive correlation with relative humidity, and a negative correlation with wind speed.
Abstract:
The essence of ecological security focuses on ecological risk and ecological vulnerability, and the study of ecological security is the hot direction of low-carbon green development, which has important theoretical value and practical significance for optimizing landscape ecological spatial structure and maintaining ecological sustainability.?Based on the interdisciplinary principles and methods of geography, ecology, remote sensing and GIS, this paper studies the characteristics and laws of climate change and landscape pattern, the coupling relationship between climate change and ecological process, landscape pattern and functional characteristics.?The relationship between landscape pattern, climate factors and soil, hydrological processes and vegetation processes in the ecosystem was analyzed, and the effects of climate change on carbon storage, NPP and ecosystem services were studied to reveal the impact of climate change on ecological processes.?The comprehensive analysis shows that climate factors affect the characteristics and changes of ecological security.Meanwhile,landscape pattern is the basis of restricting regional ecological security, and the effects of climate change and landscape pattern on ecological security mechanism are complex.
The essence of ecological security focuses on ecological risk and ecological vulnerability, and the study of ecological security is the hot direction of low-carbon green development, which has important theoretical value and practical significance for optimizing landscape ecological spatial structure and maintaining ecological sustainability.?Based on the interdisciplinary principles and methods of geography, ecology, remote sensing and GIS, this paper studies the characteristics and laws of climate change and landscape pattern, the coupling relationship between climate change and ecological process, landscape pattern and functional characteristics.?The relationship between landscape pattern, climate factors and soil, hydrological processes and vegetation processes in the ecosystem was analyzed, and the effects of climate change on carbon storage, NPP and ecosystem services were studied to reveal the impact of climate change on ecological processes.?The comprehensive analysis shows that climate factors affect the characteristics and changes of ecological security.Meanwhile,landscape pattern is the basis of restricting regional ecological security, and the effects of climate change and landscape pattern on ecological security mechanism are complex.
Abstract:
In order to solve the problems of low water-body extraction accuracy, poor multi-scale feature capture ability, high model complexity of Convolutional Neural Network (CNN) in complex background image, a Receptive Field Attention LinkNet (FRA-LinkNet) is proposed. FRA-LinkNet is a high-resolution optical remote sensing image water-body extraction model, which is a combination of Receptive Field Block (RFB) and Channel Attention Block (CAB) based on LinkNet. Firstly, RFB is used to obtain high-level water-body semantic information and multi-scale feature map; Secondly, CAB is added to weighted aggregate encoding and decoding features, to suppress background features and enhance water-body semantics. Compared with state-of-the-art CNN model, the proposed RFA-LinkNet has high performance, robustness, and precision water-body extraction.
In order to solve the problems of low water-body extraction accuracy, poor multi-scale feature capture ability, high model complexity of Convolutional Neural Network (CNN) in complex background image, a Receptive Field Attention LinkNet (FRA-LinkNet) is proposed. FRA-LinkNet is a high-resolution optical remote sensing image water-body extraction model, which is a combination of Receptive Field Block (RFB) and Channel Attention Block (CAB) based on LinkNet. Firstly, RFB is used to obtain high-level water-body semantic information and multi-scale feature map; Secondly, CAB is added to weighted aggregate encoding and decoding features, to suppress background features and enhance water-body semantics. Compared with state-of-the-art CNN model, the proposed RFA-LinkNet has high performance, robustness, and precision water-body extraction.
Abstract:
The process of screening and developing new drugs through experiments is very slow and requires a lot of manpower and material resources, and the use of computer-aided prediction of the molecular properties of drugs can greatly save the time and cost of drug development. Therefore, in order to enable anti-breast cancer candidate drugs to have good biological activity and ADMET properties for inhibiting ERα, for the collected 1974 compounds, the random forest classifier was first used to screen the top 20 molecular descriptions with the most significant effects on biological activity. The QSAR model is established using this and pIC50 value as characteristic data. Secondly, based on the PSO optimized BP neural network to predict the biological activity values of 50 new compounds, the model fit is 0.8337, and the root mean square error is 0.7315, which is more realistic than the predicted value of the BP neural network before optimization. Subsequently, in order to improve the success rate of drug development, the ADMET classification prediction model was constructed using PSO to optimize the SVM based on the existing ADMET property data. The algorithm cross-validation CV accuracy rate reached 94.0767%, and the prediction accuracy rates of the five index models were all above 79%. The results show that the established model has better prediction performance than the benchmark model, and the adopted prediction strategy is effective, which can provide a reference for the research and development of anti-breast cancer drugs.
The process of screening and developing new drugs through experiments is very slow and requires a lot of manpower and material resources, and the use of computer-aided prediction of the molecular properties of drugs can greatly save the time and cost of drug development. Therefore, in order to enable anti-breast cancer candidate drugs to have good biological activity and ADMET properties for inhibiting ERα, for the collected 1974 compounds, the random forest classifier was first used to screen the top 20 molecular descriptions with the most significant effects on biological activity. The QSAR model is established using this and pIC50 value as characteristic data. Secondly, based on the PSO optimized BP neural network to predict the biological activity values of 50 new compounds, the model fit is 0.8337, and the root mean square error is 0.7315, which is more realistic than the predicted value of the BP neural network before optimization. Subsequently, in order to improve the success rate of drug development, the ADMET classification prediction model was constructed using PSO to optimize the SVM based on the existing ADMET property data. The algorithm cross-validation CV accuracy rate reached 94.0767%, and the prediction accuracy rates of the five index models were all above 79%. The results show that the established model has better prediction performance than the benchmark model, and the adopted prediction strategy is effective, which can provide a reference for the research and development of anti-breast cancer drugs.
Abstract:
In this paper, an impedance composite control based on disturbance observer is proposed for manipulator system with unknown disturbances. For tracking the desired second order impedance dynamic model, the composite control strategy is designed, which includes disturbance observer (DOB), the impedance controller and the position controller. First, disturbance observer is used to estimate the unknown disturbances; then impedance controller corrects input angle; then position controller tracks the adjusted angle; finally, this composite control strategy ensures the impedance error can converge to a small neighborhood. The desired dynamic tracking of the second order impedance model is realized. The effectiveness of the proposed control method is proved by simulation examples.
In this paper, an impedance composite control based on disturbance observer is proposed for manipulator system with unknown disturbances. For tracking the desired second order impedance dynamic model, the composite control strategy is designed, which includes disturbance observer (DOB), the impedance controller and the position controller. First, disturbance observer is used to estimate the unknown disturbances; then impedance controller corrects input angle; then position controller tracks the adjusted angle; finally, this composite control strategy ensures the impedance error can converge to a small neighborhood. The desired dynamic tracking of the second order impedance model is realized. The effectiveness of the proposed control method is proved by simulation examples.
Abstract:
Phase-locked loop (PLL) with a pre-filtering stage is a powerful tool to study the grid synchronization technology. However, the dynamic performance of this class of commonly used PLL (e.g., second-order generalized integrator PLL, complex-coefficient filter PLL) is constrained by the low cut-off frequency of the front-stage structure. Thus, a three-phase PLL technique based on fractional-order generalized integrator is proposed. The front-stage filter of the PLL is composed of fractional-order integrators, which can generate two diagonal signals whose phase difference is 45?. Through the correlation linear operation, the positive- and negative-sequence components of the grid voltage can be extracted from the diagonal signals. Combined with the post-stage synchronous rotating frame PLL, a mathematical model of the entire PLL system is built, and the third order optimal design method is used to correct the system and determine the related control parameters. The study shows that, the cut-off frequency of the fractional-order generalized integrator is obviously higher than that of the second-order generalized integrator, which is helpful to improve the dynamic quality of the PLL system. The simulation and experimental results show that the dynamic performance of the proposed PLL is better than that of the multiple second-order generalized integrator PLL.
Phase-locked loop (PLL) with a pre-filtering stage is a powerful tool to study the grid synchronization technology. However, the dynamic performance of this class of commonly used PLL (e.g., second-order generalized integrator PLL, complex-coefficient filter PLL) is constrained by the low cut-off frequency of the front-stage structure. Thus, a three-phase PLL technique based on fractional-order generalized integrator is proposed. The front-stage filter of the PLL is composed of fractional-order integrators, which can generate two diagonal signals whose phase difference is 45?. Through the correlation linear operation, the positive- and negative-sequence components of the grid voltage can be extracted from the diagonal signals. Combined with the post-stage synchronous rotating frame PLL, a mathematical model of the entire PLL system is built, and the third order optimal design method is used to correct the system and determine the related control parameters. The study shows that, the cut-off frequency of the fractional-order generalized integrator is obviously higher than that of the second-order generalized integrator, which is helpful to improve the dynamic quality of the PLL system. The simulation and experimental results show that the dynamic performance of the proposed PLL is better than that of the multiple second-order generalized integrator PLL.
Abstract:
The efficiency of models is becoming more and more important in computer vision. In this paper, by studying the neural network structure for flame detection, several key optimization schemes are proposed to improve the efficiency of the model and the detection effect. First, this paper proposes a backbone network (FIRE-Net) constructed from a multi-convolution combined structure, which can efficiently extract rich flame features from multiple scales; second, by using an improved weighted bidirectional feature Pyramid network (BiFPN-mini) to quickly achieve multi-scale feature fusion; third, a new attention mechanism (FIRE-Attention) is proposed to make the detector more sensitive to flame characteristics. based on the above optimization , Developed a new flame detector FIRE-DET, which can always achieve faster detection efficiency than existing deep learning-based detection methods under the condition of limited hardware resources. The FIRE-DET model was trained on a self-built dataset, and the final flame recognition accuracy and frame rate reached 97% and 85FPS, respectively. The experimental results show that this flame detection model detection performance is better than the mainstream algorithm. This article provides a more common solution to solve the problem of flame detection.
The efficiency of models is becoming more and more important in computer vision. In this paper, by studying the neural network structure for flame detection, several key optimization schemes are proposed to improve the efficiency of the model and the detection effect. First, this paper proposes a backbone network (FIRE-Net) constructed from a multi-convolution combined structure, which can efficiently extract rich flame features from multiple scales; second, by using an improved weighted bidirectional feature Pyramid network (BiFPN-mini) to quickly achieve multi-scale feature fusion; third, a new attention mechanism (FIRE-Attention) is proposed to make the detector more sensitive to flame characteristics. based on the above optimization , Developed a new flame detector FIRE-DET, which can always achieve faster detection efficiency than existing deep learning-based detection methods under the condition of limited hardware resources. The FIRE-DET model was trained on a self-built dataset, and the final flame recognition accuracy and frame rate reached 97% and 85FPS, respectively. The experimental results show that this flame detection model detection performance is better than the mainstream algorithm. This article provides a more common solution to solve the problem of flame detection.
Abstract:
Atmospheric weighted mean temperature (Tm) plays a key role in GNSS atmospheric precipitable water vapor retrieval. In this paper, for the poor applicability of the existing Tm models in Tibetan Plateau, a Tm model considering surface temperature, altitude, latitude and seasonal variation, named as TPTm, is established using the observation data of 13 radiosonde stations from 2014 to 2017 in Tibetan Plateau. Then, the TPTm model is assessed by comparing with the widely used Bevis model, the local refined Bevis model (Bevis-TP model) and GPT2w model using the radiosonde data in 2018 as reference values. The results show that: Compared with Bevis, Bevis-TP, GPT2w-5 (5° resolution) and GPT2w-1 (1° resolution) models, the TPTm model has better performance, and the Root Mean Square (RMS) errors of the TPTm model is improved by 54.5%, 30.8%, 36.3% and 27.6%, respectively, with the annual bias and RMS error of 0.07 K and 2.76 K, respectively. In addition, the TPTm model has theoretical RMSPWV and RMSPWV/PWV values of 0.10 mm and 1.02 %, respectively, when used to estimate GNSS-PWV. Therefore, the TPTm model has critical applications in GNSS-PWV retrieval in Tibetan Plateau.
Atmospheric weighted mean temperature (Tm) plays a key role in GNSS atmospheric precipitable water vapor retrieval. In this paper, for the poor applicability of the existing Tm models in Tibetan Plateau, a Tm model considering surface temperature, altitude, latitude and seasonal variation, named as TPTm, is established using the observation data of 13 radiosonde stations from 2014 to 2017 in Tibetan Plateau. Then, the TPTm model is assessed by comparing with the widely used Bevis model, the local refined Bevis model (Bevis-TP model) and GPT2w model using the radiosonde data in 2018 as reference values. The results show that: Compared with Bevis, Bevis-TP, GPT2w-5 (5° resolution) and GPT2w-1 (1° resolution) models, the TPTm model has better performance, and the Root Mean Square (RMS) errors of the TPTm model is improved by 54.5%, 30.8%, 36.3% and 27.6%, respectively, with the annual bias and RMS error of 0.07 K and 2.76 K, respectively. In addition, the TPTm model has theoretical RMSPWV and RMSPWV/PWV values of 0.10 mm and 1.02 %, respectively, when used to estimate GNSS-PWV. Therefore, the TPTm model has critical applications in GNSS-PWV retrieval in Tibetan Plateau.
Abstract:
Predicting residential energy consumption is tantamount to forecasting a multivariate time series. A specific window for several sensor signals can induce various features extracted to forecast the energy consumption by using a predic-tion model. However, it is still a challenging task because of irregular patterns inside including hidden correlations between power attributes. In order to extract the complicated irregular energy patterns and selectively learn the spati-otemporal features to reduce the translational variance between energy attributes, we propose a deep learning model based on the multi-headed attention with the convolutional recurrent neural network. Compared with the simple time series model, the model uses convolution and weighting mechanism to model the local correlation between power attributes and active power. It exploits the attention scores calculated with softmax and dot product operation in the network to model the transient and impulsive nature of energy demand, predicting the instantaneous pulse power consumption effectively. Experiments with the dataset of University of California, Irvine (UCI) household electric power consumption consisting of a total 2,075,259 time-series show that the proposed model reduces the prediction error by 31.01% compared to the state-of-the-art deep learning model. Especially, the multi-headed attention im-proves the prediction performance even more by up to 27.91% than the single-attention.
Predicting residential energy consumption is tantamount to forecasting a multivariate time series. A specific window for several sensor signals can induce various features extracted to forecast the energy consumption by using a predic-tion model. However, it is still a challenging task because of irregular patterns inside including hidden correlations between power attributes. In order to extract the complicated irregular energy patterns and selectively learn the spati-otemporal features to reduce the translational variance between energy attributes, we propose a deep learning model based on the multi-headed attention with the convolutional recurrent neural network. Compared with the simple time series model, the model uses convolution and weighting mechanism to model the local correlation between power attributes and active power. It exploits the attention scores calculated with softmax and dot product operation in the network to model the transient and impulsive nature of energy demand, predicting the instantaneous pulse power consumption effectively. Experiments with the dataset of University of California, Irvine (UCI) household electric power consumption consisting of a total 2,075,259 time-series show that the proposed model reduces the prediction error by 31.01% compared to the state-of-the-art deep learning model. Especially, the multi-headed attention im-proves the prediction performance even more by up to 27.91% than the single-attention.
Abstract:
The rapid growth of malicious applications has brought a huge security threat to mobile intelligent terminals. It is of great significance to achieve high-precision detection of malicious applications for mobile network information security. In view of this, this paper proposes a malicious application detection method based on improved deep residual shrinkage network. Firstly, the traffic features are preprocessed into convolutional neural network inputs, and then the channel attention mechanism and spatial attention mechanism are introduced to weight the sample features from the channel and spatial dimensions. Then, the deep residual shrinkage network is introduced to adaptively filter out the redundant features of the samples, and the parameters are back propagated through the identical connection optimization, so as to reduce the difficulty of model training and classification, and finally realize the high-precision identification of Android malicious applications. The proposed method can avoid manual feature extraction, achieve high-precision classification and has certain generalization ability. Experimental results show that the accuracy of the proposed method is 99.40%, 99.95% and 97.33% in 2-classification, 4 -classification and 42-classification of malicious applications, which is 0.21%, 2.65% and 25.85% higher than the existing methods.Experimental results show that the accuracy of the proposed method in 2-classification, 4-classification and 42-classification of malicious applications is 99.40%, 99.95% and 97.33% respectively. Compared with the existing methods, the proposed method has higher classification performance and generalization ability.
The rapid growth of malicious applications has brought a huge security threat to mobile intelligent terminals. It is of great significance to achieve high-precision detection of malicious applications for mobile network information security. In view of this, this paper proposes a malicious application detection method based on improved deep residual shrinkage network. Firstly, the traffic features are preprocessed into convolutional neural network inputs, and then the channel attention mechanism and spatial attention mechanism are introduced to weight the sample features from the channel and spatial dimensions. Then, the deep residual shrinkage network is introduced to adaptively filter out the redundant features of the samples, and the parameters are back propagated through the identical connection optimization, so as to reduce the difficulty of model training and classification, and finally realize the high-precision identification of Android malicious applications. The proposed method can avoid manual feature extraction, achieve high-precision classification and has certain generalization ability. Experimental results show that the accuracy of the proposed method is 99.40%, 99.95% and 97.33% in 2-classification, 4 -classification and 42-classification of malicious applications, which is 0.21%, 2.65% and 25.85% higher than the existing methods.Experimental results show that the accuracy of the proposed method in 2-classification, 4-classification and 42-classification of malicious applications is 99.40%, 99.95% and 97.33% respectively. Compared with the existing methods, the proposed method has higher classification performance and generalization ability.
Abstract:
The evaluation of ecological benefits of urban renewal based on sustainable development is receiving increasing attention. Based on the practice of urban renewal in China, this paper construct a Structural Equation Model (SEM) of citizen satisfaction, takes Xi'an happiness forest belt urban renewal project as an example to evaluate urban renewal ecological benefits. By using six indicators, namely, air quality, heat island effect, vegetation greenery, resource conservation, urban transportation and urban space to measure the ecological benefits of urban renewal, it evaluates citizens' satisfaction with the ecological benefits of Xi'an happiness forest belt urban renewal project. The study found out that most of the citizens’ expectation can be met as the achievement rates of ecological expectation and overall expectation of citizens for Xi'an happiness forest belt urban renewal project were 94.3% and 97.4% respectively. Citizen expectation has a positive impact on citizen perception of ecological benefits, and the path coefficient is 0.381; citizen perception of ecological benefits has a significant positive impact on citizen satisfaction, and the path coefficient is 0.903; and citizen satisfaction has a significant positive impact on citizen trust, and the path coefficient is 0.955. Based on the research conclusions, the future urban renewal should continue to improve the quality of ecological services to improve residents' ecological environment experience, promote the establishment of a coordinated development mechanism of ecological, economic, social and cultural benefits of urban renewal, and establish a coordinated construction and governance mechanism of multiple subjects of government, residents and developers of urban renewal.
The evaluation of ecological benefits of urban renewal based on sustainable development is receiving increasing attention. Based on the practice of urban renewal in China, this paper construct a Structural Equation Model (SEM) of citizen satisfaction, takes Xi'an happiness forest belt urban renewal project as an example to evaluate urban renewal ecological benefits. By using six indicators, namely, air quality, heat island effect, vegetation greenery, resource conservation, urban transportation and urban space to measure the ecological benefits of urban renewal, it evaluates citizens' satisfaction with the ecological benefits of Xi'an happiness forest belt urban renewal project. The study found out that most of the citizens’ expectation can be met as the achievement rates of ecological expectation and overall expectation of citizens for Xi'an happiness forest belt urban renewal project were 94.3% and 97.4% respectively. Citizen expectation has a positive impact on citizen perception of ecological benefits, and the path coefficient is 0.381; citizen perception of ecological benefits has a significant positive impact on citizen satisfaction, and the path coefficient is 0.903; and citizen satisfaction has a significant positive impact on citizen trust, and the path coefficient is 0.955. Based on the research conclusions, the future urban renewal should continue to improve the quality of ecological services to improve residents' ecological environment experience, promote the establishment of a coordinated development mechanism of ecological, economic, social and cultural benefits of urban renewal, and establish a coordinated construction and governance mechanism of multiple subjects of government, residents and developers of urban renewal.
Abstract:
Virtual reality haptic rendering has higher requirements for image texture feature extraction. Texture factors are complex and irregular. A single texture extraction algorithm can not accurately describe the characteristics of image texture. Therefore, a texture material classification algorithm based on GLCM and Tamura fusion features is proposed. In addition, this paper optimizes the GLCM and proposes T-GLCM operator, which improves the rotation invariance of GLCM pair and reduces a lot of redundant information. In this paper, the Tamura texture features are used to quantify the image, and then the feature regions are quantized and cascaded into a set of feature vectors. The texture features of T-GLCM are fused, and the texture materials are classified by support vector machine (SVM). The experimental results show that the algorithm has higher classification accuracy and is better than the traditional texture feature extraction algorithm.
Virtual reality haptic rendering has higher requirements for image texture feature extraction. Texture factors are complex and irregular. A single texture extraction algorithm can not accurately describe the characteristics of image texture. Therefore, a texture material classification algorithm based on GLCM and Tamura fusion features is proposed. In addition, this paper optimizes the GLCM and proposes T-GLCM operator, which improves the rotation invariance of GLCM pair and reduces a lot of redundant information. In this paper, the Tamura texture features are used to quantify the image, and then the feature regions are quantized and cascaded into a set of feature vectors. The texture features of T-GLCM are fused, and the texture materials are classified by support vector machine (SVM). The experimental results show that the algorithm has higher classification accuracy and is better than the traditional texture feature extraction algorithm.
Abstract:
Lane detection plays an important role in the field of intelligent transportation. Its detection accuracy and speed have an important impact on assisted driving and automatic driving. Aiming at the problems of poor accuracy and slow speed of lane line recognition by deep learning methods, an efficient lane line segmentation method LaneSegNet is proposed. Firstly, based on the principle of encoding and decoding network, a backbone network Lane net is constructed to extract the lane line feature information and segment the lane line; Then, the multi-scale divided convolution feature fusion network is used to greatly expand the receptive field of the model and extract the global feature information; Finally, hybrid attention network is used to obtain rich lane line features and enhance the information related to the current task. The experimental results show that the accuracy of this method is 97.6% in TuSimple data set; On the CULane dataset, the detection accuracy of this method on standard pavement is 92.5%, and the comprehensive detection accuracy of multiple pavement is 75.2%. The LaneSegNet lane line detection method proposed in this paper has better segmentation accuracy and reasoning speed than other models, and has stronger adaptability and robustness.
Lane detection plays an important role in the field of intelligent transportation. Its detection accuracy and speed have an important impact on assisted driving and automatic driving. Aiming at the problems of poor accuracy and slow speed of lane line recognition by deep learning methods, an efficient lane line segmentation method LaneSegNet is proposed. Firstly, based on the principle of encoding and decoding network, a backbone network Lane net is constructed to extract the lane line feature information and segment the lane line; Then, the multi-scale divided convolution feature fusion network is used to greatly expand the receptive field of the model and extract the global feature information; Finally, hybrid attention network is used to obtain rich lane line features and enhance the information related to the current task. The experimental results show that the accuracy of this method is 97.6% in TuSimple data set; On the CULane dataset, the detection accuracy of this method on standard pavement is 92.5%, and the comprehensive detection accuracy of multiple pavement is 75.2%. The LaneSegNet lane line detection method proposed in this paper has better segmentation accuracy and reasoning speed than other models, and has stronger adaptability and robustness.
Abstract:
In the target tracking, the tracking efficiency is effectively improved after introducing regularization in the tracking algorithm based on the correlation filter, but it spends a lot of energy adjusting the predefined parameters. In addition, the target response occurs in non-target areas will lead to tracking drift. Therefore, a context-aware correlation filter algorithm based on time and automatic spatial regularization is proposed. First, during the tracking process, the automatic spatial regularization is realized using the target local response change, adding the automatic space regularization module to the target function to enable the filter to focus on the learning of the target object. Second, the tracker makes full use of the global context information of the target, combined with the automatic spatial regularization, to enable the filter to learn more information about the target in time, and reduce the impact of the background on the tracking performance. Subsequently, we add a temporal awareness to the filter to fully learn the change of targets between adjacent frames to obtain more accurate model samples. Experimental results show that the algorithm has better tracking effect in distance accuracy and success rate when comparing to other tracking algorithms.
In the target tracking, the tracking efficiency is effectively improved after introducing regularization in the tracking algorithm based on the correlation filter, but it spends a lot of energy adjusting the predefined parameters. In addition, the target response occurs in non-target areas will lead to tracking drift. Therefore, a context-aware correlation filter algorithm based on time and automatic spatial regularization is proposed. First, during the tracking process, the automatic spatial regularization is realized using the target local response change, adding the automatic space regularization module to the target function to enable the filter to focus on the learning of the target object. Second, the tracker makes full use of the global context information of the target, combined with the automatic spatial regularization, to enable the filter to learn more information about the target in time, and reduce the impact of the background on the tracking performance. Subsequently, we add a temporal awareness to the filter to fully learn the change of targets between adjacent frames to obtain more accurate model samples. Experimental results show that the algorithm has better tracking effect in distance accuracy and success rate when comparing to other tracking algorithms.
Abstract:
In this study, the kinetic energy equation is firstly derived from the basic equation in the rotating coordinate system, and then the energy functional describing the change rate of tropical cyclone (TC) intensity is defined by the local change rate of kinetic energy. Afterwards, the Euler-Lagrange equation is obtained by taking the variation of the functional. The equation shows that when the change rate of TC intensity reaches the maximum, the friction, the pressure gradient force, the gravity and the gradient of kinetic energy are balanced. Therefore, the vector determined by these four forces in the balance equation can be used as a predictor of TC intensity, which can more accurately determine the time when the intensity change rate of TCs reaches the maximum. Furthermore, the maximum vortex in a rotational field is extracted by the variational decomposition of wind field, and the analytical solution of vorticity and flow field are obtained when TC intensity changes the fastest. The conclusion has certain practical value for studying the variation trend of velocity and the spatial structure of TC in the evolution of TC, and provides certain theoretical guidance for TC track and intensity forecast.
In this study, the kinetic energy equation is firstly derived from the basic equation in the rotating coordinate system, and then the energy functional describing the change rate of tropical cyclone (TC) intensity is defined by the local change rate of kinetic energy. Afterwards, the Euler-Lagrange equation is obtained by taking the variation of the functional. The equation shows that when the change rate of TC intensity reaches the maximum, the friction, the pressure gradient force, the gravity and the gradient of kinetic energy are balanced. Therefore, the vector determined by these four forces in the balance equation can be used as a predictor of TC intensity, which can more accurately determine the time when the intensity change rate of TCs reaches the maximum. Furthermore, the maximum vortex in a rotational field is extracted by the variational decomposition of wind field, and the analytical solution of vorticity and flow field are obtained when TC intensity changes the fastest. The conclusion has certain practical value for studying the variation trend of velocity and the spatial structure of TC in the evolution of TC, and provides certain theoretical guidance for TC track and intensity forecast.
Abstract:
Urban growth and contraction are the inevitable result of rapid urbanization. Excessive growth or contraction will bring problems such as resource mismatch and ecological environment deterioration. Based on the nighttime light data of China's districts and counties from 1992 to 2019, this paper deeply explores the temporal and spatial dynamic characteristics and causes of urban growth and contraction in China. In order to provide reference for the future new urbanization construction, balanced allocation of resources and national development policy-making. The results show that: ① Urban growth and contraction alternate in China, and the polarization between growth and contraction is more and more obvious. ② The center of urban growth has been moving in Nanyang city. The growth structure is mainly zonal and supplemented by a dot. The belt growth structure includes the coastal urban belt and the recently formed "Beijing―Zhengzhou―Nanchang" urban belt. Point structure refers to the agglomeration growth phenomenon with inland provincial capital cities as the core. ③ The center of contraction shifted from Xingtai City to Ankang City, and the center of contraction changed obviously. The contraction structure gradually changed from scattered point contraction to "Beijing―Xi'an―Chengdu―Kunming" belted contraction structure. ④ Global financial change cycle, industrial structure transformation, national policies and other factors are the main causes of urban growth and contraction in China.
Urban growth and contraction are the inevitable result of rapid urbanization. Excessive growth or contraction will bring problems such as resource mismatch and ecological environment deterioration. Based on the nighttime light data of China's districts and counties from 1992 to 2019, this paper deeply explores the temporal and spatial dynamic characteristics and causes of urban growth and contraction in China. In order to provide reference for the future new urbanization construction, balanced allocation of resources and national development policy-making. The results show that: ① Urban growth and contraction alternate in China, and the polarization between growth and contraction is more and more obvious. ② The center of urban growth has been moving in Nanyang city. The growth structure is mainly zonal and supplemented by a dot. The belt growth structure includes the coastal urban belt and the recently formed "Beijing―Zhengzhou―Nanchang" urban belt. Point structure refers to the agglomeration growth phenomenon with inland provincial capital cities as the core. ③ The center of contraction shifted from Xingtai City to Ankang City, and the center of contraction changed obviously. The contraction structure gradually changed from scattered point contraction to "Beijing―Xi'an―Chengdu―Kunming" belted contraction structure. ④ Global financial change cycle, industrial structure transformation, national policies and other factors are the main causes of urban growth and contraction in China.
Abstract:
As an important method of face modeling, 3D deformation model (3DMM) has been widely used in 3D modeling, image synthesis and other fields. Due to the influence of training data type, quantity, linear basis and other factors, 3DMM has the phenomenon of over-constraint and cannot provide enough flexibility to represent high-frequency deformation. In order to further improve the problem, the three-dimensional deformation basis function is embedded into the deep neu-ral network, which provides a new idea for improving the representation ability of the model. In order to improve the efficiency of network learning, a two-path neural network is constructed to achieve the balance between global path and local path. By improving nonlinear 3DMM in both learning objectives and network structure, we propose a model that can capture higher levels of detail than linear or previous nonlinear models. The algorithm is compared with the simula-tion experiment, which shows that the algorithm has lower normalized average error in 3D face modeling, and the gen-erated 3D face model has good robustness and accurate reconstruction, and achieves better 3D face reconstruction per-formance.
As an important method of face modeling, 3D deformation model (3DMM) has been widely used in 3D modeling, image synthesis and other fields. Due to the influence of training data type, quantity, linear basis and other factors, 3DMM has the phenomenon of over-constraint and cannot provide enough flexibility to represent high-frequency deformation. In order to further improve the problem, the three-dimensional deformation basis function is embedded into the deep neu-ral network, which provides a new idea for improving the representation ability of the model. In order to improve the efficiency of network learning, a two-path neural network is constructed to achieve the balance between global path and local path. By improving nonlinear 3DMM in both learning objectives and network structure, we propose a model that can capture higher levels of detail than linear or previous nonlinear models. The algorithm is compared with the simula-tion experiment, which shows that the algorithm has lower normalized average error in 3D face modeling, and the gen-erated 3D face model has good robustness and accurate reconstruction, and achieves better 3D face reconstruction per-formance.
Abstract:
Sea breeze front (SBF) is landward edge of the of the sea breeze circulation, which has the similar characteristics to the cold front. Because SBF can trigger strong convection and cause disastrous weather such as thunderstorms in coastal areas, it is the key object of coastal meteorological research. With the development of numerical models and computer technology, numerical simulation has become an important method in meteorological research. Therefore, the numerical simulation of SBF is attracting more and more attention. In the last 40 years, the numerical studies of SBF have been continuously developed in China. To deepen the understanding of SBF and do better research in the future, this paper summarizes the numerical simulation of SBF in China and focuses on the effects of environmental flow, topography, urban heat island and atmosphere stability on SBF. In addition, the influence of the physical parameterization on SBF and the work of large eddy simulation are summarized. Some issues are addressed for the further study.
Sea breeze front (SBF) is landward edge of the of the sea breeze circulation, which has the similar characteristics to the cold front. Because SBF can trigger strong convection and cause disastrous weather such as thunderstorms in coastal areas, it is the key object of coastal meteorological research. With the development of numerical models and computer technology, numerical simulation has become an important method in meteorological research. Therefore, the numerical simulation of SBF is attracting more and more attention. In the last 40 years, the numerical studies of SBF have been continuously developed in China. To deepen the understanding of SBF and do better research in the future, this paper summarizes the numerical simulation of SBF in China and focuses on the effects of environmental flow, topography, urban heat island and atmosphere stability on SBF. In addition, the influence of the physical parameterization on SBF and the work of large eddy simulation are summarized. Some issues are addressed for the further study.
Abstract:
Federated learning is a kind of distributed machine learning technology that ensures that local data is not compromised when training with big data for machine learning models. However, a series of attacks shows that the adversary can steal private information from machine learning model parameters even if local data is inaccessible. Thus, from the intermediate model parameters transmission between the participants and the aggregator in the training phase to the finally released aggregated model, there are many privacy threats during the model release process of federated learning. As a result, many privacy-preserving federated learning approaches have emerged, primarily based on cryptography and differential privacy technology. This paper surveys the various privacy threats and adversary models that may appear when we publish local models and the aggregated model of federated learning. Furthermore, we systematically summarize the related defense technologies and research results. Additionally, we also look forward to the development of privacy-preserving federated learning.
Federated learning is a kind of distributed machine learning technology that ensures that local data is not compromised when training with big data for machine learning models. However, a series of attacks shows that the adversary can steal private information from machine learning model parameters even if local data is inaccessible. Thus, from the intermediate model parameters transmission between the participants and the aggregator in the training phase to the finally released aggregated model, there are many privacy threats during the model release process of federated learning. As a result, many privacy-preserving federated learning approaches have emerged, primarily based on cryptography and differential privacy technology. This paper surveys the various privacy threats and adversary models that may appear when we publish local models and the aggregated model of federated learning. Furthermore, we systematically summarize the related defense technologies and research results. Additionally, we also look forward to the development of privacy-preserving federated learning.
Abstract:
In this paper, the problem of cooperative state feedback control for a class of large-scale linear systems described by Delta operator is studied. Firstly, according to the independent subsystem described by Delta operator, an interconnected closed-loop control system is given through cooperative state feedback controller. Then, based on linear matrix inequality techniques, the sufficient conditions for the design of coordinated state feedback stabilizing controller and cooperative state feedback guaranteed cost controller are given, and the validity of proposed method is proved by Lyapunov stability theory. Finally, simulation examples show the effectiveness and superiority of the proposed algorithm.
In this paper, the problem of cooperative state feedback control for a class of large-scale linear systems described by Delta operator is studied. Firstly, according to the independent subsystem described by Delta operator, an interconnected closed-loop control system is given through cooperative state feedback controller. Then, based on linear matrix inequality techniques, the sufficient conditions for the design of coordinated state feedback stabilizing controller and cooperative state feedback guaranteed cost controller are given, and the validity of proposed method is proved by Lyapunov stability theory. Finally, simulation examples show the effectiveness and superiority of the proposed algorithm.
Abstract:
Traditional navigation methods can only detect static obstacles on the path, but can not detect moving obstacles. Therefore, a collision avoidance path navigation method for blind people based on deep learning is proposed. The speech signal is collected, and the speech feature parameters are obtained by using the speech recognition model. According to the speech feature parameters, the content of the speech sequence input by the blind person is recognized, and the destination of the blind person is determined. The obstacle detection model is constructed to detect the shape features, moving direction and speed of the obstacles on the blind's location and their destination path, and calculate the distance between the initial position and the arrival position. The convolution neural network in deep learning is used to plan the optimal collision avoidance path and realize the blind collision avoidance path navigation. The experimental results show that the radial velocity of the obstacles detected by this method is almost the same on the x-axis and y-axis, and the moving direction and velocity of the obstacles can be tracked and monitored in real time. The speed obtained by this method is basically consistent with the actual speed of the obstacle, and the error is between 0.2 and 0.4cm/s. when the test time is 50min, the obstacle avoidance accuracy reaches 96.5%, which can achieve the optimal collision avoidance path planning and navigation.
Traditional navigation methods can only detect static obstacles on the path, but can not detect moving obstacles. Therefore, a collision avoidance path navigation method for blind people based on deep learning is proposed. The speech signal is collected, and the speech feature parameters are obtained by using the speech recognition model. According to the speech feature parameters, the content of the speech sequence input by the blind person is recognized, and the destination of the blind person is determined. The obstacle detection model is constructed to detect the shape features, moving direction and speed of the obstacles on the blind's location and their destination path, and calculate the distance between the initial position and the arrival position. The convolution neural network in deep learning is used to plan the optimal collision avoidance path and realize the blind collision avoidance path navigation. The experimental results show that the radial velocity of the obstacles detected by this method is almost the same on the x-axis and y-axis, and the moving direction and velocity of the obstacles can be tracked and monitored in real time. The speed obtained by this method is basically consistent with the actual speed of the obstacle, and the error is between 0.2 and 0.4cm/s. when the test time is 50min, the obstacle avoidance accuracy reaches 96.5%, which can achieve the optimal collision avoidance path planning and navigation.
Select AllDeselectExport
Display Method:
Abstract:
With the rapid development of internet of things,cloud computing,and mobile internet,the rise of Big Data has attracted more and more concern,which brings not only great benefits but also crucial challenges on how to manage and utilize Big Data better.This paper describes the main aspects of Big Data including definition,data sources,key technologies,data processing tools and applications,discusses the relationship between Big Data and cloud computing,internet of things and mobile internet technology.Furthermore,the paper analyzes the core technologies of Big Data,Big Data solutions from industrial circles,and discusses the application of Big Data.Finally,the general development trend on Big Data is summarized.The review on Big Data is helpful to understand the current development status of Big Data,and provides references to scientifically utilize key technologies of Big Data.
With the rapid development of internet of things,cloud computing,and mobile internet,the rise of Big Data has attracted more and more concern,which brings not only great benefits but also crucial challenges on how to manage and utilize Big Data better.This paper describes the main aspects of Big Data including definition,data sources,key technologies,data processing tools and applications,discusses the relationship between Big Data and cloud computing,internet of things and mobile internet technology.Furthermore,the paper analyzes the core technologies of Big Data,Big Data solutions from industrial circles,and discusses the application of Big Data.Finally,the general development trend on Big Data is summarized.The review on Big Data is helpful to understand the current development status of Big Data,and provides references to scientifically utilize key technologies of Big Data.
Abstract:
根据个人学习研究稳定性的心得体会，首先介绍了前苏联伟大的数学力学家Ｌｙａｐｕｎｏｖ院士的博士论文《运动稳定性的一般问题》在全世界产生的超过１个世纪的巨大影响．叙述了由该博士论文首创的几个巨大成就何以能奠定１门学科的基础，从而开创了１个新的重要的研究方向，以及留给后人很多很多研究的课题的理由．特别地，用事实和科学断语回答了“Ｌｙａｐｕｎｏｖ稳定性已领风骚１００多年，余晖还几何”的问题．明确表明１个观点：稳定性将是１个“永恒的主题”，不老的科学，定将永恒地给人启迪，洞察力，智慧和思想．
根据个人学习研究稳定性的心得体会，首先介绍了前苏联伟大的数学力学家Ｌｙａｐｕｎｏｖ院士的博士论文《运动稳定性的一般问题》在全世界产生的超过１个世纪的巨大影响．叙述了由该博士论文首创的几个巨大成就何以能奠定１门学科的基础，从而开创了１个新的重要的研究方向，以及留给后人很多很多研究的课题的理由．特别地，用事实和科学断语回答了“Ｌｙａｐｕｎｏｖ稳定性已领风骚１００多年，余晖还几何”的问题．明确表明１个观点：稳定性将是１个“永恒的主题”，不老的科学，定将永恒地给人启迪，洞察力，智慧和思想．
Abstract:
System identification is the theory and methods of establishing mathematical models of systems.The mathematical modeling has a long research history,but the system identification discipline has only several tens of years.In this short decades,system identification has achieved great developments,new identification methods are born one after another,and the research results cover the theory and applications of natural science and social sciences,including physics,biology,earth science,meteorology,computer science,economics,psychology,political science and so on.In this context,we come back to ponder some basic problems of system identification,which is not without benefits for the development of system identification.This is a paper of an introduction to system identification which briefly introduces the definition of identification,system models and identification models,the basic steps and purposes of identification,including the experimental design of identification and data preprocessing,and the types of identification methods,including the least squares identification methods,gradient identification methods,auxiliary model based identification methods,and multi innovation identification methods,and hierarchical identification methods,etc
System identification is the theory and methods of establishing mathematical models of systems.The mathematical modeling has a long research history,but the system identification discipline has only several tens of years.In this short decades,system identification has achieved great developments,new identification methods are born one after another,and the research results cover the theory and applications of natural science and social sciences,including physics,biology,earth science,meteorology,computer science,economics,psychology,political science and so on.In this context,we come back to ponder some basic problems of system identification,which is not without benefits for the development of system identification.This is a paper of an introduction to system identification which briefly introduces the definition of identification,system models and identification models,the basic steps and purposes of identification,including the experimental design of identification and data preprocessing,and the types of identification methods,including the least squares identification methods,gradient identification methods,auxiliary model based identification methods,and multi innovation identification methods,and hierarchical identification methods,etc
Abstract:
Recently,coordinated control of multi-agent systems has been a hot topic in the control field,due to its wide application in cooperative control of multiple autonomous vehicles,traffic control of vehicles,formation control of unmanned aircrafts,resource allocation in networks and so on.Firstly,the introduction of background about multi-agent systems,the concepts of agents and the knowledge of the graph theory has been given.And then research status of swarming/flocking problems,formation control problems,consensus problems and network optimization are summarized and analyzed at home and abroad,including coordination control of multi-agent systems.Finally,some problems about multi-agent systems to be solved in future are proposed,in order to urge deep study on the theory and application in coordinated control of multi-agent systems.
Recently,coordinated control of multi-agent systems has been a hot topic in the control field,due to its wide application in cooperative control of multiple autonomous vehicles,traffic control of vehicles,formation control of unmanned aircrafts,resource allocation in networks and so on.Firstly,the introduction of background about multi-agent systems,the concepts of agents and the knowledge of the graph theory has been given.And then research status of swarming/flocking problems,formation control problems,consensus problems and network optimization are summarized and analyzed at home and abroad,including coordination control of multi-agent systems.Finally,some problems about multi-agent systems to be solved in future are proposed,in order to urge deep study on the theory and application in coordinated control of multi-agent systems.
Abstract:
设计了一个三维声源定位系统，提出了一个新的系统模型，并对传统的基于声波到达时间差TDOA的算法进行了优化。通过检测麦克风接收到信号的时间差，结合已知的阵列元的空间位置确定声源的位置。该系统声源采集部分由4个阵列成正四面体的麦克风组成。算法的硬件实现由TMS320C5416DSP芯片完成'整个系统实现了声源定位的功能。
设计了一个三维声源定位系统，提出了一个新的系统模型，并对传统的基于声波到达时间差TDOA的算法进行了优化。通过检测麦克风接收到信号的时间差，结合已知的阵列元的空间位置确定声源的位置。该系统声源采集部分由4个阵列成正四面体的麦克风组成。算法的硬件实现由TMS320C5416DSP芯片完成'整个系统实现了声源定位的功能。
Abstract:
In recent years,cloud computing as a new computing service model has become a research hotspot in computer science.This paper is to give a brief analysis and survey on the current cloud computing systems from the definition,deployment model,characteristics and key technologies.Then,the major international and domestic research enterprises and application products on cloud computing are compared and analyzed.Finally,the challenges and opportunities in current research of cloud computing are discussed,and the future directions are pointed out.So,it will help to provide a scientific analysis and references for use and operation of cloud computing.
In recent years,cloud computing as a new computing service model has become a research hotspot in computer science.This paper is to give a brief analysis and survey on the current cloud computing systems from the definition,deployment model,characteristics and key technologies.Then,the major international and domestic research enterprises and application products on cloud computing are compared and analyzed.Finally,the challenges and opportunities in current research of cloud computing are discussed,and the future directions are pointed out.So,it will help to provide a scientific analysis and references for use and operation of cloud computing.
2017,9(2):159-167, DOI: 10.13878/j.cnki.jnuist.2017.02.006
Abstract:
Various indoor positioning techniques have been developed and widely applied in both manufacturing processes and people's lives.Due to the electromagnetic interference and multipath effects,traditional Wi-Fi,Bluetooth and other wireless locating technologies are difficult to achieve high accuracy.Modulated white LED can provide both illumination and location information to achieve highly accurate indoor positioning.In this paper,we first introduce several modulation methods of visible light positioning systems and compare the characteristics of different modulation methods.Then,we propose a viable indoor positioning scheme based on visible light communications and discuss two different demodulation methods.In the following,we introduce several positioning algorithms used in visible light communication system.Finally,the problems and prospects of the visible light communication based indoor positioning are discussed.
Various indoor positioning techniques have been developed and widely applied in both manufacturing processes and people's lives.Due to the electromagnetic interference and multipath effects,traditional Wi-Fi,Bluetooth and other wireless locating technologies are difficult to achieve high accuracy.Modulated white LED can provide both illumination and location information to achieve highly accurate indoor positioning.In this paper,we first introduce several modulation methods of visible light positioning systems and compare the characteristics of different modulation methods.Then,we propose a viable indoor positioning scheme based on visible light communications and discuss two different demodulation methods.In the following,we introduce several positioning algorithms used in visible light communication system.Finally,the problems and prospects of the visible light communication based indoor positioning are discussed.
2017,9(2):174-178, DOI: 10.13878/j.cnki.jnuist.2017.02.008
Abstract:
With the deepening study of nonlinear effect in optical fiber,the distributed optical fiber sensor has been widely studied and applied.In this paper,the application of optical fiber sensor is introduced.To realize different types of fiber distributed sensing,the principle of three kinds of scattered light based on Brillouin scattering,Raman scattering,and Rayleigh scattering is summarized.Finally,the future development direction of fiber distributed sensing is prospected.
With the deepening study of nonlinear effect in optical fiber,the distributed optical fiber sensor has been widely studied and applied.In this paper,the application of optical fiber sensor is introduced.To realize different types of fiber distributed sensing,the principle of three kinds of scattered light based on Brillouin scattering,Raman scattering,and Rayleigh scattering is summarized.Finally,the future development direction of fiber distributed sensing is prospected.
Abstract:
We propose a scheme to produce continuous-variable(CV) pair-entanglement frequency comb by nondegenerate optical parametric down-conversion in an optical oscillator cavity in which a multichannel variational period poled LiTaO3 locates as a gain crystal.Using the CV entanglement criteria,we prove that every pair generated from the corresponding channel is entangled.The characteristics of signal and idler entanglement are discussed.The CV pair-entanglement frequency comb may be very significant for the application in quantum communication and computation networks.
We propose a scheme to produce continuous-variable(CV) pair-entanglement frequency comb by nondegenerate optical parametric down-conversion in an optical oscillator cavity in which a multichannel variational period poled LiTaO3 locates as a gain crystal.Using the CV entanglement criteria,we prove that every pair generated from the corresponding channel is entangled.The characteristics of signal and idler entanglement are discussed.The CV pair-entanglement frequency comb may be very significant for the application in quantum communication and computation networks.
Abstract:
On account of the power quality signal under stable state,this paper integrates the function of Hanning window with Fast Fourier Transform(FFT),and uses it to harmonic analysis for power quality.Matlab simulation is carried out for the feasibility of the proposed windowed FFT method,and results show that the integration of Hanning window function with FFT can significantly reduce the harmonic leakage,effectively weaken the interference between the harmonics,and accurately measure the amplitude and phase of power signal.
On account of the power quality signal under stable state,this paper integrates the function of Hanning window with Fast Fourier Transform(FFT),and uses it to harmonic analysis for power quality.Matlab simulation is carried out for the feasibility of the proposed windowed FFT method,and results show that the integration of Hanning window function with FFT can significantly reduce the harmonic leakage,effectively weaken the interference between the harmonics,and accurately measure the amplitude and phase of power signal.
Abstract:
As a modulation with relatively strong anti-interference capacity,quadrature phase shift keying(QPSK) has been extensively used in wireless satellite communication.This paper describes the Matlab simulation of QPSK demodulation,and designs an all-digital QPSK demodulation with FPGA.The core of demodulation is synchronization,which includes carrier synchronization and signal synchronization.The carrier synchronization is completed through numerical Costas loop,while signal synchronization through modulus square spectrum analysis,and the results are simulated on Matlab.The communication functions are implemented by upgradable or substitutable softwares as many as possible,based on the idea of software radio communication.The parameter values through Matlab simulation,combined with appropriate hardware system,technically realize the design of the proposed all-digital meteorological satellite demodulator based on FPGA.
As a modulation with relatively strong anti-interference capacity,quadrature phase shift keying(QPSK) has been extensively used in wireless satellite communication.This paper describes the Matlab simulation of QPSK demodulation,and designs an all-digital QPSK demodulation with FPGA.The core of demodulation is synchronization,which includes carrier synchronization and signal synchronization.The carrier synchronization is completed through numerical Costas loop,while signal synchronization through modulus square spectrum analysis,and the results are simulated on Matlab.The communication functions are implemented by upgradable or substitutable softwares as many as possible,based on the idea of software radio communication.The parameter values through Matlab simulation,combined with appropriate hardware system,technically realize the design of the proposed all-digital meteorological satellite demodulator based on FPGA.
Abstract:
This paper proposes a two-step detection scheme that begins thick and ends thin,to mine the outliers of multivariable time series (MTS).According to the confidence interval of the data in sliding window,characteristics of both variation trend value and relevant variation trend value were constructed,which were then used in the two detection processes.Meanwhile,the rapid extraction algorithm for characteristics is studied.The outlier detection scheme is then applied to mine outliers before and after an accident happened at a 110 kV Grid Transformer Substation in Jiangsu province.Data sets of various equipment tables,which were collected by OPEN3000 data surveillance system,were checked by the proposed detection scheme,and experiment result indicates that this algorithm can rapidly and precisely locate the outliers.
This paper proposes a two-step detection scheme that begins thick and ends thin,to mine the outliers of multivariable time series (MTS).According to the confidence interval of the data in sliding window,characteristics of both variation trend value and relevant variation trend value were constructed,which were then used in the two detection processes.Meanwhile,the rapid extraction algorithm for characteristics is studied.The outlier detection scheme is then applied to mine outliers before and after an accident happened at a 110 kV Grid Transformer Substation in Jiangsu province.Data sets of various equipment tables,which were collected by OPEN3000 data surveillance system,were checked by the proposed detection scheme,and experiment result indicates that this algorithm can rapidly and precisely locate the outliers.
Abstract:
In order to solve the sudoku more efficiently,a novel approach was proposed.We employed the real number coding to get rid of the integer constraint,meanwhile used the L0 norm to guarantee the sparsity of the solution.Moreover,the L1 norm was used to approximate the L0 norm on the basis of RIP and KGG condition.Finally,the slack vectors were introduced to transfer the L1 norm into a convex linear programming problem,which was solved by the primal dual interior point method.Experiments demonstrate that this algorithm reach 100％ success rate on easy,medium,difficult,and evil levels,and reach 864％ success rate on only 17 clue sudokus.Besides,the average computation time is quite short,and has nothing to do with the difficulty of sudoku itself.In all,this algorithm is superior to both constraint programming and Sinkhorn algorithm in terms of success rate and computation time.
In order to solve the sudoku more efficiently,a novel approach was proposed.We employed the real number coding to get rid of the integer constraint,meanwhile used the L0 norm to guarantee the sparsity of the solution.Moreover,the L1 norm was used to approximate the L0 norm on the basis of RIP and KGG condition.Finally,the slack vectors were introduced to transfer the L1 norm into a convex linear programming problem,which was solved by the primal dual interior point method.Experiments demonstrate that this algorithm reach 100％ success rate on easy,medium,difficult,and evil levels,and reach 864％ success rate on only 17 clue sudokus.Besides,the average computation time is quite short,and has nothing to do with the difficulty of sudoku itself.In all,this algorithm is superior to both constraint programming and Sinkhorn algorithm in terms of success rate and computation time.
2017,9(6):575-582, DOI: 10.13878/j.cnki.jnuist.2017.06.002
Abstract:
Knowledge graph technology is widely concerned and studied during recent years,in this paper we introduce the construction methods,recent development of knowledge graph in details,we also summarize the interdisciplinary applications of knowledge graph and future directions of research.This paper details the key technologies of textual,visual and multi-modal knowledge graph,such as information extraction,knowledge fusion and knowledge representation.As an important part of the knowledge engineering,knowledge graph,especially the development of multi-modal knowledge graph,is of great significance for efficient knowledge management,knowledge acquisition and knowledge sharing in the era of big data.
Knowledge graph technology is widely concerned and studied during recent years,in this paper we introduce the construction methods,recent development of knowledge graph in details,we also summarize the interdisciplinary applications of knowledge graph and future directions of research.This paper details the key technologies of textual,visual and multi-modal knowledge graph,such as information extraction,knowledge fusion and knowledge representation.As an important part of the knowledge engineering,knowledge graph,especially the development of multi-modal knowledge graph,is of great significance for efficient knowledge management,knowledge acquisition and knowledge sharing in the era of big data.
Abstract:
With the continuous increase of road vehicles,occasional congestion caused by traffic accidents seriously affect the commuting efficiency of traveler and the overall operation level of road network.Real-time and exact forecasting of short-term traffic flow volume is the key point to intelligent traffic system and precondition to solve the congestion situation by route guidance and clearing.According to the uncertain and non-linear features of traffic volume,a model integrated of the improved BP neural network and autoregressive integrated moving average (ARIMA) model is established to forecast the short-term traffic flow.The case application result shows that the combined model has an advantage over the single models in forecasting performance and forecasting accuracy.
With the continuous increase of road vehicles,occasional congestion caused by traffic accidents seriously affect the commuting efficiency of traveler and the overall operation level of road network.Real-time and exact forecasting of short-term traffic flow volume is the key point to intelligent traffic system and precondition to solve the congestion situation by route guidance and clearing.According to the uncertain and non-linear features of traffic volume,a model integrated of the improved BP neural network and autoregressive integrated moving average (ARIMA) model is established to forecast the short-term traffic flow.The case application result shows that the combined model has an advantage over the single models in forecasting performance and forecasting accuracy.
Abstract:
Based upon GDAS and GBL NCEP reanalysis data with resolution 1°×1°and 2.5°×2.5°respectively,the trajectory of the air mass at 100 m altitude over Hetian meteorological station is simulated by HYSPLIT (Hybrid Single Particle Lagrangian Integrated Trajectory Model),which is developed by Air Source Laboratory of NOAA,to estimate the effect of integration error and resolution error on the trajectory calculation error.The contribution of the integration error is found to be very small,which increases slightly with the integration time length and has no relation to the resolution of the meteorological data.The resolution error varies at different time point,and is found to be related to the topography,the weather system and the interpolation.The simulated trajectories using datasets with different resolution differed with each other significantly,indicating that the resolution error contributes more to the trajectory calculation error than calculation error.
Based upon GDAS and GBL NCEP reanalysis data with resolution 1°×1°and 2.5°×2.5°respectively,the trajectory of the air mass at 100 m altitude over Hetian meteorological station is simulated by HYSPLIT (Hybrid Single Particle Lagrangian Integrated Trajectory Model),which is developed by Air Source Laboratory of NOAA,to estimate the effect of integration error and resolution error on the trajectory calculation error.The contribution of the integration error is found to be very small,which increases slightly with the integration time length and has no relation to the resolution of the meteorological data.The resolution error varies at different time point,and is found to be related to the topography,the weather system and the interpolation.The simulated trajectories using datasets with different resolution differed with each other significantly,indicating that the resolution error contributes more to the trajectory calculation error than calculation error.
DownloadsMore+
Authors CornerMore+
Search
Ranking List
- 1A novel sudoku solving method based on sparse optimization
- 2Analysis of summer aerosol characteristics in various functional areas of Nanjing
- 3The time series mixture model and its application in international oil’s price prediction
- 4Facility location model
- 5ICIST 2011 Call for Papers
- 6Talking on the theory, methods and application of Lyapunov stability
- 7Vision detection of driving fatigue based on lip chroma Fisher classifier
- 83-D sound localization based on four microphone array
- 9Big Data:Conceptions,key technologies and application
- 10System identifcation.Part A:Introduction to the identification
External Links
Address：No. 219, Ningliu Road, Nanjing, Jiangsu Province
Postcode：210044
Phone：025-58731025
