Accurate identification of line parameters is crucial for the stable operation and optimization of power grids. With the rapid advancement of artificial intelligence

deep learning-based methods for power grid line parameter identification have demonstrated significant advantages in effectiveness and robustness. However

these methods often overlook historical trends and topological relationships of network branches

resulting in models that fail to fully learn critical spatiotemporal information

thereby decreasing parameter identification accuracy. To address this

we propose a dynamic spatiotemporal adaptive graph neural network-based method for power grid line parameter identification. It utilizes the maximum information coefficient and Bayesian optimization based on the tree-structured Parzen estimator to automatically select the most relevant input measurement features while adjusting model hyperparameters

and constructs a spatiotemporal graph dataset based on historical branch features and topological information. The method employs graph convolutional networks and temporal convolutional networks to extract line features

enhanced by a dynamic spatiotemporal adaptive module to capture each line’s unique characteristics. In case studies on the IEEE 39-bus system

the method shows improved accuracy and robustness against measurement noise

data loss

and topology changes compared to existing algorithms.