Topology Identification Based on GCN-Graph Lasso

Traditional topology identification algorithms usually depend on state monitoring, and in practice, sensors may cause data synchronization issues, which could be misidentified as topology anomalies. To this end, this paper first uses a Graph Convolutional Network to extract the latent correlation between the collected grid data, and then inputs the obtained weight matrix into Graph Lasso to predict the precision matrix. An adaptive threshold filter is applied to the precision matrix to obtain the predicted connection relationships. Finally, the proposed method is validated on the IEEE 33-node and 123-node systems. The test results show that the proposed method can accurately detect and locate topological changes, contributing to the safe and reliable operation of the power system.