Abstract: In order to improve the transient assessment model’s ability to extract electrical dynamic features and its generalization ability when facing changes in system topology, this paper proposes an online assessment model with a temporal and spatial dual-channel parallel structure. Firstly, the dynamic information of the transient time series data of the model is captured by the Gated Recurrent Unit (GRU). The nonlinear fitting relationship between power system topology and transient stability state is constructed based on the Graph Attention Network (GAT). And the spatial and temporal features of the two channels are fused through the attention mechanism to obtain more reliable evaluation results. Then, when the topology of the original system changes, the model is combined with transfer learning technology to update the network parameters of the model and realize the online update of the model. Finally, through the simulation and verification of IEEE 39-node system and IEEE 300-node system, the model evaluation accuracy reaches 98.62% and 98.51%, respectively. The results show that the proposed method can realize efficient transient stability evaluation and has strong robustness.

Key words: electric power system, transient stability assessment, deep learning, spatiotemporal feature, attention feature fusion, transfer learning