Multi-view Dynamic Spatiotemporal Graph Network for Improved Communication Fault Location in Smart Substation

Introduction: Accurately locating faults in smart substation communication networks is critical for safety. Current methods, focusing on static timestamps, neglect fault spatiotemporal evolution, hindering dynamic state capture and early prevention. This study proposes a novel fault location model to address this gap. Methods: We develop a multi-view spatiotemporal dynamic graph neural network model. It constructs multi-view graphs from network topology and attributes. A multi-view module captures spatial dependencies among fault nodes via cross-view contrastive learning. A gated recurrent unit with a temporal dynamic window adaptively extracts state evolution trends at each timestamp, integrating spatiotemporal information to enhance early fault perception. Results: Evaluated on a 220kV smart substation communication network, the proposed model demonstrates significantly higher fault location accuracy compared to existing models. Discussion: The model successfully captures the spatiotemporal evolution characteristics ignored by previous approaches. While effective, computational complexity requires future optimization for broader deployment. Conclusion: This model provides a more accurate solution for fault location in smart substation communication networks by effectively modeling spatiotemporal dynamics, offering potential for improved reliability and early fault intervention.