Power System Transient Stability Assessment Based on Graph Convolutional Network

Fast and accurate transient stability assessment (TSA) is significant for power system. Topological information is often ignored by some machine learning methods, which may have a profound influence for TSA. For the purpose of improving the assessment accuracy and the adaptability to topology changes for online TSA, graph convolutional network (GCN) is introduced. GCN can aggregate node features and topological information. A jumping knowledge layer is added to prevent overfitting by concatenating the output of every GCN layer. Steady-state features of power system that are got through state estimation from the SCADA system are employed as the input. Tests on IEEE 39-Bus system indicate that the GCN model has a higher performance than some other methods. Several system topology changes are considered, and good adaptability to different topologies is indicated.