基于GRA-Optuna-LSTM模型的泵站前池水位多步预测研究OA

To improve the accuracy of water level forecasting of the pump station forebay and extend the forecasting horizon,this paper takes the Ningxia Guhai Irrigation Cascade Pump Station as the research object and proposes an LSTM prediction model that integrates Grey Relational Analysis(GRA)and the Optuna hyperparameter optimization method.Through GRA,four key influencing factors were identified:the flow rate of the Kuosan pump station,the water level in the outflow pool of the Kuoer pump station,the flow rate of the Kuoer pump station,and the flow difference between the two pump stations.Optuna was used to automatically optimize the LSTM hyperparameters,and the prediction results under different forecasting horizons(1,2,3 and 4 h)were compared with LSTM,Optuna-XGBoost,and Optuna-BP models.The results show that the Optuna-LSTM model has the lowest error metrics,significantly outperforming the comparative models,demonstrating superior accuracy and generalization capability,thus providing a high-precision and highly generalizable data-driven solution for short-term forecasting of pump station water levels.