基于XGBoost-RF-StackGA模型的作物需水量多步预测OA

To improve agricultural water resources allocation and irrigation scheduling in arid regions,we developed an XGBoost-RF-StackGA hybrid prediction model.To address the nonlinear and multi-scale characteristics of crop water requirement series,this model integrates eXtreme gradient boosting(XG-Boost),random forest(RF),and stacking regression,with genetic algorithm(GA).This study focuses on the prediction of crop water requirements of spring wheat,maize,and rice in the Hanyan Canal irriga-tion area.Grey relational analysis(GRA)and XGBoost-SHAP were combined to identify key meteoro-logical factors,genetic algorithm(GA)was used to optimize the hyperparameters of XGBoost and RF.On this basis,multi-step prediction models for crop water requirements over future 1,3,7,10,and 15 days were established and then compared with the NGBoost-RF-StackGA model.The results showed that the proposed model achieved higher accuracy and stronger stability across different crops and predic-tion durations.For short-term prediction,the coefficient of determination(R2)ranged from 0.981 to 0.991,and the Nash-Sutcliffe efficiency coefficient(NSE)exceeded 0.977.For medium and long-term prediction,the R2 and NSE remained above 0.876 and 0.821,respectively.These results indicate that the proposed model can effectively capture the complex nonlinear relationships between meteorologi-cal factors and crop water requirements.The findings can provide technical support for multi-temporal-scale crop water requirement prediction and refined agricultural water management in the irrigation areas along the Yellow River.