基于多模型融合的中长期径流集成预测方法  

Integrated Forecasting Method for Medium to Long-term Runoff Based on Multi-model Fusion

作　　者：朱非林[1] 陈嘉乙 张咪 徐向荣 钟平安[1] ZHU Feilin;CHEN Jiayi;ZHANG Mi;XU Xiangrong;ZHONG Ping'an(College of Hydrology and Water Resources,Hohai University,Nanjing 210024,Jiangsu,China)

机构地区：[1]河海大学水文水资源学院,江苏南京210024

出　　处：《水力发电》2024年第2期6-13,29,共9页Water Power

基　　金：国家重点研发计划项目(2022YFC3202801);国家自然科学基金资助项目(52009029)。

摘　　要：中长期水文预报是流域水资源规划与合理配置的重要依据。为提高中长期径流预测精度,提出了一种基于多模型融合的水库中长期径流集成预测方法。该方法将ARMA、BP、LSTM、RF和SVR等5个异质预测模型进行融合,同时采用超参数优化方法确定各模型的最优参数。将其用于青海省龙羊峡水库的中长期径流预报中,结果表明,通过Stacking融合算法建立的集成预测模型相较于单一模型,取得了更高的预测精度(R2值由0.71提升至0.82)。此方法可为提升流域中长期径流预测精度提供一定参考。The medium to long-term hydrological forecasting is a crucial basis for watershed water resources planning and rational allocation.In order to improve the accuracy of medium to long-term hydrological forecasting,a medium to long-term runoff integrated forecasting method based on multi-model fusion is proposed for reservoirs,This method combines five heterogeneous prediction models of ARMA,BP,LSTM,RF and SVR,and at the same time,a hyperparameter optimization method is employed to determine the optimal parameters for each model.The proposed method is applied to the medium to long-term runoff forecasting of Longyangxia Reservoir in Qinghai Province,and the results demonstrate that the ensemble prediction model established through the Stacking fusion algorithm achieves higher prediction accuracy compared to individual model with the R value increasing from 0.71 to 0.82.The method presented in this paper can provide valuable reference for improving the accuracy of medium to long-term runoff prediction in watersheds.

关 键 词：中长期径流预报 ARMA BP LSTM RF SVR 多模型融合 集成预测 Stacking融合算法 超参数寻优 龙羊峡水库 

分 类 号：TV124[水利工程—水文学及水资源]