Substation flood precipitation forecasting model based on spatio-temporal sequence UNet depth network

Extreme climate change can intensify precipitation events in local areas. In order to predict the future precipitation intensity in the area around the substation within a short period of time, a forecasting model based on a spatio-temporal sequence depth network is proposed in this paper. The precipitation forecast is formulated as a spatio-temporal sequence forecasting problem, in which both the input and forecast targets are spatio-temporal sequences. The input-to-state and state-to-state transformations are performed using the convolutional structure in Unet, and an end-to-end trainable model is built for the precipitation forecasting problem. Experiments conducted on radar images from 2019 to 2022 show that the proposed method can better capture spatio-temporal correlations and outperforms fully connected long- and short-term memory networks and convolutional neural networks for short-time precipitation forecasting in areas around substations.