Research on flood risk prediction of substation based on Temporal Fusion Transformer

The flood disaster caused by extreme weather seriously affects the safety and stability of substations, and flood risk prediction is the key to ensure the reliability of power supply. Flooding in substations is affected by many factors, which involve dynamic and static features, making its accurate prediction challenging. To address these challenges, this paper proposes a substation flood risk prediction model based on Temporal Fusion Transformer (TFT). Different modules are used for feature importance analysis, static covariate feature extraction, and temporal feature mining of risk sequences. The results of experiments and case studies demonstrate that this model is superior to the current mainstream time series prediction models, and can be used for multi-risk scenarios analysis, thereby providing auxiliary decision support for flood prevention efforts.