Fast Resilience Assessment for Power Systems Under Typhoons Based on Spatial Temporal Graphs

Despite great progress in modeling the resilience response of power systems under extreme events, it remains difficult to assess the evolutionary trend of system performance at a specific observation moment during such events. Conventional simulation-based assessment methods are typically time-consuming because a series of scenario-specific optimization problems must be solved as a prerequisite. Thus, a spatial-temporal graph-based approach is proposed for fast resilience assessment to provide timely warning information. The key factors, including observable meteorological information, component vulnerabilities, emergency dispatch, and repair strategies, are modeled in the form of matrices that depict the spatial and temporal relationships. Based on these matrices, a spatiotemporal graph neural network is developed to fit the mapping relationship between observable states and resilience indicators, which is trained offline and enables fast assessment via forward inference. Regarding the uncertainties of various extreme scenarios, the evaluation procedure combines the whole-process simulation and single-state replay technologies, which can respectively consider the uncertainties and provide deterministic data labeling for assessment. Finally, the effectiveness of the proposed method is verified on the benchmarks, including the IEEE 118-bus system and a realistic 2868-bus system.