STGCN-Based Spatio-Temporal Evolution Modeling of Distribution System Disaster Chains

Extreme weather events such as typhoons can cause severe damage to distribution systems through complex disaster-chain effects. Previous studies in disaster modeling for power systems have made important progress but often emphasize isolated events or rely on static features, which may limit their ability to capture the dynamic spatio-temporal interactions underlying disaster evolution. To address this limitation, this paper proposes a Spatio-Temporal Graph Convolutional Network (STGCN)-based framework for modeling the evolution of typhoon disaster chains in distribution systems. Based on the causal relationships among primary and secondary disaster events, the STGCN is hierarchically decomposed into multiple sub-STGCN modules to reduce computational complexity and enhance interpretability. Using the IEEE 33-bus system as a case study, the proposed approach is compared with conventional models including CNN, LSTM, and MLP. The results show that the improved STGCN achieves the lowest mean squared error (MSE) on spatially correlated datasets, demonstrating its superior capability to extract spatial dependencies and characterize the spatio-temporal dynamics of disaster evolution in distribution systems.