Intelligent emergency control for power systems via spatio-temporal graph learning

As the interconnection of power grids deepens and renewable penetration rises, transient stability assessment (TSA) and emergency control become increasingly challenging due to higher system uncertainty and complexity. Accurate and timely TSA is essential for the reliable execution of generator tripping schemes. This paper proposes a novel framework for transient stability evaluation and emergency generator tripping based on spatiotemporal graph deep learning. A wide-area transient response feature is constructed using a structure-preserving transient energy function, and the resulting feature sequences are modeled as spatiotemporal graphs. A hybrid model combining Convolutional Neural Networks (CNN) and Graph Attention Networks (GAT) is developed to perform TSA and predict coherent generator clusters. The CNN module captures temporal dependencies, while the GAT module learns spatial relationships across the power network. Based on the predicted coherency clusters, a generator tripping scheme is derived using the one-machine infinite-bus (OMIB) equivalence method. Compared with previous work [1], which applied graph learning for coherency identification, our method further integrates temporal dynamics into both TSA and emergency control using an end-to-end spatiotemporal learning framework. The proposed method is validated on the IEEE 39-bus system and real-world grid scenarios, demonstrating superior accuracy, adaptability, and potential for real-time emergency decision support.