In the monitoring and management of power transmission systems,traditional parameter identification methods focus solely on data from individual branches,neglecting information from adjacent branches within the overall grid topology. Coupled with data contamination issues such as missing data and noise caused by external factors,the accuracy of parameter identification requires improvement. To address these issues,this paper proposes a graph neural network model with GraphSAGE (Graph Sample and Aggregate) as its core. First,the model learns the grid topology information and generates hidden-layer features through neighbor information aggregation. Then,a multi-task learning strategy is introduced for different real-world scenarios,enabling joint identification of multiple branch parameters while optimizing them. Finally,a noise filtering module is incorporated to introduce anti-noise factors into branch features,allowing the model to effectively handle missing data and noise,thereby enhancing robustness. Experimental results show that the proposed method outperforms traditional approaches in accuracy and stability.