Abstract:In practical engineering, the sampling intervals of networked control systems are often subject to undesirable physical constraints, which results in noisy sampling intervals.In view of this, we focus on the stabilization of networked control systems with noisy sampling intervals and stochastic time-varying delays.First, a closed-loop stochastic system model, whose system matrix is characterized by high nonlinearity and dual randomness, is obtained by considering the noisy sampling intervals and stochastic time-varying delays in a unified framework.In order to deal with the difficulties arising from the nonlinearity and dual randomness of the system matrices, the confluent Vandermonde matrix approach and Kronecker product operation are utilized, and then the mathematical expectations of the product of three matrices related to the system matrices are calculated.Based on this, a sufficient condition for stochastic stability of the closed-loop system is obtained, and a stabilization controller is designed by solving a linear matrix inequality.Finally, two examples are provided to verify the effectiveness of the designed method.

Abstract:Observer-based discrete Markov jump systems (D-MJSs) are random, but their faults to be estimated are deterministic signals, so it is difficult to directly study the fault estimation of the system.Here, an approach of constructing auxiliary systems is proposed, which constructs deterministic intermediate variables by taking expectations of random variables, and then the established intermediate estimator can estimate the system state and faults at the same time.In addition, based on a new scaling method, the uncertainty of the Transition Probability Matrix (TPM) is considered.All conditions are given by Linear Matrix Inequalities (LMIs), and the state error is ensured by stable input state.Numerical simulation examples are provided to prove the effectiveness and superiority of the proposed approach.

Abstract:Due to the low accuracy of speaker recognition for short-term speech or under overlapping noisy situations, a new speaker recognition algorithm based on deep learning is proposed and then deployed on an embedded device.The encoding layer and loss function are the two aspects to improve the speaker recognition system in robustness.For the encoding layer, the NeXtVLAD technique based on differential encoding is used to model both static and dynamic speaker features at frame level.For the loss function, the cosine-prototypical loss function based on small-sample learning framework is fused with the additional margin classification loss function AM-Softmax to train the speaker recognition model, which enables the model to collect similar features and separate dissimilar features as much as possible in the feature space.Then the improved speaker recognition algorithm is deployed on the Raspberry Pi platform to realize speaker recognition with fast inference.The experimental results illustrate that the system can accomplish speaker recognition in real time and accurately under various open set scenarios, and meet the requirements of practical applications.

Abstract:Here we consider a class of nonlinear neutral stochastic delay differential equations.The coefficients of the drift term and diffusion term could increase superlinearly, and the neutral term satisfies the contractive mapping condition.The truncated θ-EM method for this type of equations is established and the convergence rate is obtained.Finally, an example is given to verify the theoretical result.

Abstract:Here the input-output finite-time stability for a class of stochastic distributed parameter switched systems is studied.First, the concept of the input-output finite-time stability of system in mean square sense is given.Second, by using the piecewise Lyapunov-Krasovskii function, average dwell time, linear matrix inequality and other tools, the sufficient conditions of input-output finite-time stability are constructed.In addition, a state feedback controller is designed to ensure the input-output finite-time stability of corresponding closed-loop system, which is proved by strict mathematical derivation.Finally, two simulation examples are given to verify the effectiveness of the results obtained.

Abstract:For stochastic linear discrete time systems, a Q-learning algorithm is proposed in this paper to solve the stochastic linear quadratic optimal tracking control problem in the infinite time domain.First, it is assumed that the reference signal required for tracking is generated by the command generator, and an augmented system consisting of the original stochastic system and the reference trajectory system is established, then the optimal tracking problem is transformed into an optimal regulation problem.Second, in order to solve the optimal tracking problem online, the stochastic system is transformed into a deterministic one, the Q function of stochastic linear quadratic optimal tracking control is defined according to the augmented system, and the augmented stochastic algebraic equation is solved online without knowing the parameters of the system model.Third, the equivalence between the Q-learning algorithm and the augmented stochastic algebraic equation is proved, and the implementation steps of the Q-learning algorithm are given.Finally, a simulation example is given to illustrate the effectiveness of the proposed Q-learning algorithm.

Abstract:This paper studies the stabilization problem of stochastic singular Markov jump systems with time-varying delays by static event-triggered control.By constructing the Lyapunov-Krasovskii functional, Jensen's inequality and free weightingm atrix technology, the stochastic admissibility conditions for the systems under the condition of static event-triggered control are proposed.On this basis, the state feedback controller is designed so that the corresponding closed-loop systems satisfy regular, impulse-free and the stochastically admissible in the mean square.Finally, a numerical simulation example is provided to illustrate the correctness and effectiveness of the method proposed in this paper.

Abstract:In deep reinforcement learning, the deep Q-network algorithm seriously overestimates the action value, which degrades the performance of agents.The double deep Q-network and dueling network structure can partially alleviate the impact of overestimation, sometimes the former one even underestimate the action value.Here, a Weighted Dueling Double Deep Q-Network (WD3QN) algorithm is proposed, in which the improved double estimators and dueling network structure are combined into the deep Q-network, and the learned possible action values are weighted to produce the final action value, which can effectively reduce the estimation error.Finally, the algorithm is applied to the classical CartPole control problem on the open AI Gym platform.The simulation results show that compared with other existing algorithms, the proposed algorithm has better learning effect, convergence and training speed.

Abstract:The prediction of time series traffic is a hot issue in machine learning in recent years.It has been found that the prediction accuracy can be greatly improved by approaches of changing the network structure (such as the number of neural network layers, the number of neurons in network layers, the connection mode between network layers, as well as the application of special network layers), and selecting appropriate optimizer and loss function.Here, we propose a multi-layer LSTM (Long Short-Term Memory) algorithm, which is a single model improved on traditional LSTM algorithm, to reduce the model's complexity and improve the efficiency of machine learning.The model includes an input layer, five hidden layers, an output layer, a full connection layer, and also a dropout layer to prevent the machine learning from over-fitting.The model uses adam as optimizer, mlse as loss function, and relu as activation function.The experimental results show that the proposed model has better generalization ability compared with traditional LSTM model.

Abstract:A BP (Back-Propagation) neural network model optimized by PSO (Particle Swarm Optimization) and based on data preprocessing is proposed for sag prediction of overhead transmission lines, in order to solve the susceptibility of sag computation to measured data of temperature, wind speed, span and other parameters.For the missing data in collected database, the Synthetic Minority Oversampling Technique (SMOTE) was used to synthesize unbalanced samples.The proposed PSO-BP neural network was trained and tested by data obtained in different working environments.Experiments were carried out to verify the effectiveness of the proposed approach.The results showed that, compared with traditional BP neural network, the proposed model has a significant reduction in the relative error of sag prediction, and can accelerate the training speed as well as improve the sag prediction accuracy.

Abstract:A forgetting factor based data-driven optimal iterative learning control method is proposed for a class of nonlinear nonaffine discrete-time systems running repeatedly in finite time.First, an iterative dynamic linearization method is introduced to transform the nonlinear system into a linear input and output incremental form.Second, the main problems of the optimal iterative learning method are analyzed.To solve the problem that the control input cannot respond in time due to the accumulation effect of historical information, an adaptive forgetting factor is designed to make the method more controllable and flexible.The proposed control method is a data-driven control approach, and the design and analysis process only depends on the input and output data of the system and does not contain any explicit model information.Finally, the effectiveness of the proposed method is verified by simulation results.

Abstract:In distribution practices, the delivery task may be delayed or completed in advance, both of which will increase the distribution cost and decrease the revenue, resulting the earliness/tardiness risk.Here, the routing optimization is studied for the fourth-party logistics with consideration of earliness/tardiness risk.A mathematical model trying to minimize the distribution costs and constrain the earliness/tardiness risk is set up, then a deletion algorithm embedded harmony search is proposed.The effectiveness of the proposed model and algorithm is verified through solving cases at different scales.

Abstract:Based on the condition of Corporate Social Responsibility (CSR), we address a supply chain composed of a manufacturer and a CSR retailer considering three cases:centralized one, decentralized one, and decentralized one with Revenue Sharing (RS) contract.The research results show that, the optimal profit of the centralized supply chain decreases with the CSR level.In the decentralized case, the retailer's profit decreases and the manufacturer's profit increases along with the increase of retailer's CSR level.While in the decentralized case with RS contract, the RS coefficient affects the retailer's optimal profit, specifically, the retailer's profit decreases along with the increase of its CSR level when the RS coefficient is within certain range, while out of that range, the retailer's profit would increase along with the increase of its CSR level;however, the manufacturer's profit always increases with retailer's CSR level.The social welfare increases along with the retailer's CSR level when the CSR level is relatively low, then it would decrease with the retailer's CSR level when the CSR level is relatively high.In the end, we provide a numerical example and derive some managerial implications.

Abstract:Using renewable energy to power Road Side Units (RSUs) is a desirable alternative, since it lowers both the carbon footprint and the cost of deployment.Therefore, for renewable energy RSUs, two online distributed scheduling strategies are proposed in this paper to maximize the number of served vehicles.In the Markov chain-based scheduling strategy, the energy state of RSUs is expressed by Markov chain, and the number of served vehicles is maximized by rewarding action.While in the threshold-based scheduling strategy, the harvested energy in RSUs as well as the energy consumed by serving vehicle is calculated to select the number of served vehicles.Simulation results show that the proposed online scheduling strategies increase the capacity of serving vehicles.

Abstract:This paper proposes the adaptive coupling coordination degree and adaptive coordination interval for the population and economic system, then employs the RBF network learning algorithm to realize the warning strategy and the analysis of influencing factors for the coordinated development between population and economic system for Ningxia's Yinchuan and Shizuishan.First, the adaptive coupling coordination degree as well as the adaptive coordination interval is introduced to formulate an adaptive warning strategy for the development relationship between population and economic system.Second, a three-layer network learning model is employed to determine and analyze the key factors affecting the coupling and coordinated development between population and economic system.Finally, the approach proposed in this paper is applied to Yinchuan and Shizuishan, based on which key factors are determined and interpreted for the coupling and coordinated development between population and economic system in these two cities.

Abstract:Here, the Response Surface Method (RSM) is used to optimize the structure of a new Axial-Flux Permanent Magnet Memory Motor (AFPMMM) proposed in this paper.Main influencing parameters of the motor are preliminarily determined based on analysis of the electromagnetic performance equation and Orthogonal Experiment (OE), among which, three factors including soft magnetic ratio, number of rotor poles and air gap length, are selected as the design factors of OE.Then Electromotive Force(EMF), Total Harmonic Distortion (THD) and cogging torque are determined as the optimization factors, and the RSM is used as the optimization method.The Response Surface (RS) experiment is established by using the finite element software Ansoft Maxwell and the RS design software Design Expert.The simulation parameters and fitting curves of the motor under the combination model of different design factors are obtained.Different optimization methods are combined to meet the specific design requirements in this paper.Theoretical analysis and experimental results verify the feasibility and effectiveness of the proposed motor and optimization method.Then the optimization scheme is obtained through comparative analysis of the test data.The results show that the optimized motor not only reduces the cost of permanent magnet material, but also ensures good EMF and small axial force fluctuation.

Abstract:The rapid development of the Internet of Things further makes its data interaction vulnerable to various attacks.To ensure the security of data transmitted by UDP, the transport layer protocol of the Internet of Things, namely the DTLS (DatagramTLS) protocol, which supports the secure transmission of UDP datagrams, has been formed on the basis of the TLS protocol architecture.However, based on certificate public key cryptography, the existing DTLS protocol has disadvantages such as complex certificate management as well as high network communication overhead, thus cannot meet the secure communication requirements of resource-constrained networks such as the Internet of Things.Here, we propose an improved certificateless public key cryptographic scheme based on discrete logarithm, and design a lightweight DTLS protocol adaptable to resource-constrained networks, and then implement the protocol based on the embedded SSL library of wolfSSL.Finally, experiments are conducted to compare the DTLS protocol based on improved certificateless public key cryptography proposed in this article with the DTLS protocol based on traditional public key certificates and the DTLS protocol based on identity markers, and experimental results verify the superiority of the proposed protocol in terms of communication overhead and handshake delay.