基于Transformer的水库垂向水温预测方法研究OA

The distribution of water temperature in front of the reservoir dam is one of the key factors determining the effectiveness of stratified water intake through stoplogs in reservoirs.However,most methods for predicting vertical water temperature rely on numerical simulations,which are limited in practical engineering applications due to their low computational efficiency and extensive parameter settings.To quickly predict the vertical water temperature in front of the reservoir dam,this paper proposes a vertical water temperature prediction method based on the Transformer structure.By unifying the time series mean and variance within the sliding window to the same distribution type,this method addresses the issue of time series distribution skewness and improves the traditional models.Training and validation were conducted using vertical water temperature distribution data from Xiluodu Reservoir from January to May during 2020-2023.Finally,the errors of the traditional Transformer model and the improved model were compared.The research results show that the traditional Transformer model can simulate the vertical water temperature distribution trend well,but the simulation effect is slightly worse in the thermocline and deep water layers.The root mean square error(RMSE)of the vertical water temperature in the validation set is 0.286℃,and the mean absolute error(MAE)is 0.203℃.After unifying the time series mean and variance within the sliding window to the same distribution type,the RMSE of the vertical water temperature in the validation set of the improved model is 0.169℃,and the MAE is 0.107℃.The simulation accuracy has been significantly improved compared to the traditional model.This study provides a method for predicting vertical water temperature in front of the reservoir dam,offering a vertical water temperature prediction tool for the formulation of stratified water intake schemes through stoplogs in practical engineering.