基于集成学习的海勃湾水库出库泥沙预测研究  

Research on Sediment Discharge Prediction Based on Ensemble Learning for Haibowan Reservoir

作　　者：郭秀吉 孙龙飞 王远见[1,2] 任智慧 李新杰[1,2] GUO Xiu-ji;SUN Long-fei;WANG Yuan-jian;REN Zhi-hui;LI Xin-jie(Yellow River Institute of Hydraulic Research,YRCC,Zhengzhou 450003,China;Key Laboratory of Lower Yellow River Channel and Estuary Regulation,MWR,Zhengzhou 450003,China)

机构地区：[1]黄河水利委员会黄河水利科学研究院,河南郑州450003 [2]水利部黄河下游河道与河口治理重点实验室,河南郑州450003

出　　处：《水电能源科学》2025年第4期98-101,79,共5页Water Resources and Power

基　　金：国家重点研发计划(2021YFC3200400);国家自然科学基金项目(U2243241,U2243236,U2243601);中央级公益性科研院所基本科研业务费专项项目(HKY-JBYW-2024-13,HKY-JBYW-2024-20,HKY-JBYW-2024-21,HKY-JBYW-2024-22)。

摘　　要：针对水库排沙影响因素众多、非线性关系复杂、难以精准预测等问题,以海勃湾水库为研究对象,采用2014~2020年进出库水沙系列数据,构建了基于集成树框架的RF、XGBoost出库含沙量预测模型,并与传统KNN、DT算法进行了拟合性能对比。结果表明,入库流量Q_(in)、入库含沙量S_(in)、进出库水位差ΔZ、坝前水位差ΔH是影响海勃湾水库排沙的主要因素;不同机器学习算法应用于水库排沙预测均是有效的,相比于KNN、DT算法,集成树模型拥有更好的拟合性能;基于Boosting算法的XGBoost模型各项预测指标最佳,在应对样本随机性和非一致性方面展现出良好的拟合稳定性及泛化能力。Reservoir sediment discharge is influenced by multiple factors and has complex nonlinear relationships,making it difficult to predict accurately.In response to the above issues,taking Haibowan Reservoir as the research object,a series of water and sediment data from 2014 to 2020 were used to construct an RF and XGBoost sediment concentration prediction model based on the integrated tree framework.The fitting performance was compared with the traditional KNN and DT algorithms.The results indicate that Q_(in),S_(in),ΔZ,andΔH are the main factors affecting the sediment discharge of Haibowan Reservoir;Different machine learning algorithms are effective in predicting reservoir sediment discharge;Compared to the KNN and DT algorithms,the ensemble tree model has better fitting performance;The XGBoost model based on boosting algorithm has the best prediction index and demonstrates good fitting stability and generalization ability in dealing with sample randomness and inconsistency.

关 键 词：集成学习 海勃湾水库 出库含沙量 预测模型 

分 类 号：TV145[水利工程—水力学及河流动力学] TV882.1