基于机器学习的汉江流域径流模拟与时滞变化分析  

Runoff simulation and time-lag change analysis in the Han River Basin based on machine learning

作　　者：黄一凡 张翔[1,2] 邓梁堃 李宜伦 刘浩源 HUANG Yifan;ZHANG Xiang;DENG Liangkun;LI Yilun;LIU Haoyuan(State Key Laboratory of Water Resources Engineering and Management,Wuhan University,Wuhan 430072,China;Hubei Key Laboratory of Water System Science for Sponge City Construction,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学水资源工程与调度全国重点实验室,湖北武汉430072 [2]武汉大学海绵城市建设水系统科学湖北省重点实验室,湖北武汉430072

出　　处：《水资源保护》2024年第6期173-180,共8页Water Resources Protection

基　　金：国家自然科学基金区域创新发展联合基金重点支持项目(U21A2002)。

摘　　要：针对传统水文模型存在参数率定困难、容易陷入局部最优解以及适用性不强等问题,基于机器学习原理构建了基于贝叶斯优化算法的长短期记忆(BOLSTM)模型,并将其用于汉江流域径流模拟;基于模拟结果,采用SHAP方法对降雨径流过程中的影响因子进行了归因分析,并采用时滞分析方法量化了南水北调中线工程对流域径流过程的影响。结果表明:BOLSTM模型的径流模拟效果较好;南水北调中线工程的建设推迟了降雨对汉江流域出口流量产生影响的时间,降雨6 d后流域出口流量有所增加;而工程建设前则会在降雨5 d后导致流域出口流量发生变化,且工程建设前汉江流域降雨对流域出口流量的影响更大。In response to the issues of parameter calibration difficulties,susceptibility to local optima,and poor applicability in traditional hydrological models,a Bayesian optimization algorithm-based long short-term memory(BOLSTM)model was constructed based on machine learning principles,and it was applied to runoff simulation in the Han River Basin based on the simulation results,the SHAP method was used to attribute the influencing factors in the rainfall-runoff process,and the time-lag analysis method was used to quantify the impact of the Middle Route of the South-to-North Water Diversion Project on the runoff process of the watershed.The results show that the runoff simulation effect of the BOLSTM model is better;the construction of the Middle Route of the South-to-North Water Diversion Project has delayed the impact of rainfall on the flow at the outlet of the Han River Basin,with the flow at the outlet of the basin increasing after 6 days of a rainfall;while before the construction of the project,the occurrence of a rainfall would lead to changes in the flow at outlet of the basin after 5 days,and the impact of rainfall in the Han River Basin on the flow at the outlet of the basin was greater before that of the project.

关 键 词：径流模拟 长短期记忆网络 机器学习 贝叶斯优化算法 时滞分析 汉江流域 

分 类 号：TV121[水利工程—水文学及水资源] P333[天文地球—水文科学]