A Data-Centric Artificial Intelligence -Based Method for Enhancing Power System Resilience under Multimodal Meteorological Coupling

With the changing global climate, power grids in southeastern coastal regions, such as Guangdong, are increasingly susceptible to the variable local weather conditions, leading to a growing number of issues like conductor galloping and tower tilting. To address these challenges, this paper proposes an optimal strategy for predicting power grid load reduction and reinforcing the grid structure under complex weather conditions. The proposed method is centered around DCAI (Data-Driven Computational Intelligence) technology, which is capable of sensitively capturing the impact of multiple compound weather conditions on power grid load loss and identifying the optimal reinforcement schemes. The method consists of three main components: (1)constructing a comprehensive meteorological intensity index based on compound weather conditions; (2) establishing a spatiotemporal mapping relationship between weather and line faults using Inverse Distance Weighting (IDW) interpolation and the Haversine distance formula; (3)developing a spatiotemporal graph neural network (ST-GNN) with physical constraints, which employs spatial encoding, temporal encoding, and physical constraint layers to build the mapping relationship between line faults and load loss. Additionally, a robust optimization model considering both reinforcement cost and safe operation constraints is established, with adversarial tests conducted under regional attacks, moving storms, and multi-pulse attacks. Based on this, corresponding power grid resilience evaluation indicators are proposed, and the effectiveness of the proposed methods is verified using an improved IEEE 39-bus test system.