Intelligent substation communication network fault location method based on dynamic spatiotemporal graph association perception

INTRODUCTION: Accurately locating faults in intelligent substation communication networks is crucial for power grid safety. Existing methods fail to fully capture dynamic fault characteristic evolution and complex dependencies within network topologies OBJECTIVES: This paper aims to (1) model spatiotemporal fault features in communication networks, (2) enhance fault pattern capture through multi-view learning, and (3) improve fault location accuracy. METHODS: We propose a multi-view spatiotemporal dynamic graph network. First, a multi-view graph neural network models spatial dependencies via cross-view comparative learning using topological and attribute data. Second, a gated recurrent unit with dynamic time windows extracts temporal evolution trends, focusing on local fault patterns and short-term dependencies. RESULTS: Evaluations on a 220kV substation communication network show our method achieves higher fault location accuracy versus baselines, effectively capturing spatiotemporal fault characteristics. CONCLUSION: The proposed framework addresses dynamic fault evolution and topological dependencies, providing a robust solution for intelligent substation fault diagnosis.