基于特征优选和极端梯度提升树的水厂供水量预测模型OA

[Objective]To support the development of smart water,meet the demand for refined scheduling of water treatment plant(WTP),and improve water supply efficiency and resource utilization,it is necessary to construct an accurate hourly water supply forecasting model to optimize production scheduling.[Methods]Taking WTP A as the research object in the east China region,this paper integrated time characteristics,meteorological conditions,equipment maintenance conditions and hydraulic parameters to build a dataset with 58 initial features.The recursive feature elimination(RFE)algorithm was used to screen 16 key features,and Bayesian optimization(BO)was combined to optimize the hyperparameters of the extreme gradient boosting tree(XGBoost)model,thereby establishing an hourly water supply forecasting model.[Results]The model achieved a mean absolute percentage error(MAPE)of 4.58％and a mean absolute error(MAE)of 676.02 m3/h after BO.Verification in April 2024 showed that when scheduled based on predictions,the 1-hour reservoir water level fluctuation could be controlled within±0.3 m at 99.66％of the time periods,and the maximum daily water level deviation was only 0.77 m without manual intervention.In August 2024,the model was integrated with the intelligent scheduling system and put into online operation.By providing 1 hour to 4 hours water supply forecasting,it reduced the number of pump start-stop cycles by 47％.[Conclusion]This model has both high accuracy and practical value,which can avoid equipment loss and energy waste,and provide technical support for the optimized scheduling of WTPs.