FDRformer: A Transformer-based electrical substation flooding disaster risk forecasting model

In this paper, we propose FDRformer model, a novel Transformer-based model designed for substation flooding disaster risk prediction. The model integrates a gated feature selection network and an encoder-decoder network to effectively reduce redundant information and capture complex nonlinear relationships among multiple variables, thereby enhancing prediction accuracy. Using real-world data from an electric power research institute in central China, the FDRformer model is trained and evaluated. Experimental results show that FDRformer outperforms existing machine learning and deep learning models, achieving a $32.87 \%$ reduction in RMSE compared to conventional deep learning methods. The proposed model is highly effective in predicting flood risks, providing valuable insights for disaster mitigation and resource allocation in substation management.