GCN-LSTM Based Transient Angle Stability Assessment Method for Future Power Systems Considering Spatial-temporal Disturbance Response Characteristics

Traditional transient angle stability analysis methods do not fully consider the spatial characteristics of the network topology and the temporal characteristics of the time-series disturbance.Hence,a data-driven method is proposed in this study,combining graph convolution network and long short-term memory network(GCN-LSTM)to analyze the transient power angle sta-bility by exploring the spatiotemporal disturbance char-acteristics of future power systems with high penetration of renewable energy sources(wind and solar energy)and power electronics.The key time-series electrical state quantities are considered as the initial input feature quantities and normalized using the Z-score,whereas the network adjacency matrix is constructed according to the system network topology.The normalized feature quan-tities and network adjacency matrix were used as the inputs of the GCN to obtain the spatial features,reflecting changes in the network topology.Subsequently,the spa-tial features are inputted into the LSTM network to ob-tain the temporal features,reflecting dynamic changes in the transient power angle of the generators.Finally,the spatiotemporal features are fused through a fully con-nected network to analyze the transient power angle stability of future power systems,and the softmax activa-tion cross-entropy loss functions are used to predict the stability of the samples.The proposed transient power angle stability assessment method is tested on a 500 kV AC-DC practical power system,and the simulation results show that the proposed method could effectively mine the spatiotemporal disturbance characteristics of power sys-tems. Moreover, the proposed model has higher accuracy, higher recall rate, and shorter training and testing times than traditional transient power angle stability algo-rithms.