Research on Improved Transformer Based-Flood Risk Prediction Model for Substations

In recent years, China has faced increasingly severe natural disasters, including frequent extreme rainstorms that significantly impact the operation of power grids especially substations. Developing effective meteorological disaster risk prediction models is crucial for enhancing flood prevention and early warning capabilities within the power system. This paper introduces an improved Transformer-based flood risk prediction model for substations, tailored for emergency prevention scenarios. To address the diverse forms and importance of features in flood control data, we propose an input feature enhancement strategy incorporating a gating selection mechanism and embedded encoding. The improved model achieves MAE, MSE, and RMSE values of $0.02214,0.00079$, and 0.02823, respectively, surpassing those of existing mainstream models. Additionally, with an R2 value of 0.94479, the model demonstrates superior accuracy and effectiveness in substation flood risk prediction.