A Spatio-Temporal Attention–Enhanced LSTM Model for Critical Fault-Set Identification Under Wildfire Conditions

Abstract

Power systems are severely threatened by wildfires, which can potentially trigger N–k cascading faults and lead to large-scale blackouts. To mitigate these risks, this paper proposes a novel critical fault-set identification model. First, an LSTM-based framework is introduced to model the time-series evolution of line states under varying load levels and external wildfire conditions. Meanwhile, a spatio-temporal attention mechanism is introduced to account for both the topological connectivity among transmission lines and their temporal dependencies. This integrated model not only addresses the temporal continuity in cascading failures but also accounts for topological complexity in the grid. Experimental results show that the model achieves a high identification accuracy of 98.05% on the test set, surpassing the performance of baselines including Transformer-based and CNN-LSTM architectures. Furthermore, it demonstrates strong adaptability to different load conditions and wildfire intensities, underscoring its practical value in wildfire scenarios.