Recovery Strategy for Power Grid based on Graph Learning Under Unconventional events

The frequent occurrence of extreme events presents substantial challenges for the efficient and swift restoration of power grids after disasters. To tackle these issues, this paper introduces an enhanced few-shot learning framework that combines Meta-Learning, Graph Neural Networks, and collaborative agents to support adaptive recovery strategy development despite limited learning samples. This framework adeptly captures high-order graph features, delivering more comprehensive insights into real-time topology and power flow, even when learning samples are scarce. Simulation tests on the IEEE 39-bus benchmark system demonstrate that our approach can yield an effective recovery strategy with improved system performance and reduced computation time. These adaptive and rapid recovery strategies provide valuable guidance for bolstering the resilience of power grids in the face of extreme events.