摘要
该文引入机器学习方法优化基于物理机制的洪水模型,构建了耦合分布式水文模型——VIC(variable infiltration capacity)-机器学习水文模型,克服了因观测资料不足而使模型模拟精度过低的局限.研究发现,优化后的耦合VIC-机器学习模型能更好地模拟实测流量,NSE系数(Nash-Sutcliffe efficiency)接近0.85,解决了原有单一基于物理模型产生的过拟合问题.通过模拟2013—2021年三峡区间库首宜昌站流量序列发现,三峡区间流量变化存在增加趋势.研究将构建的模型用于2024年三峡区间内典型洪水事件(大宁河2024年1号洪水)的复盘分析,进一步验证了该模型在重现洪水事件中的有效性.
This study introduces machine learning methods to optimize a flood model based on physical mechanisms,and develops a coupled flood model incorporating the VIC model and machine learning for the Three Gorges Reservoir Region.This approach overcomes the limitation of low simulation accuracy due to insufficient observational data.Our results show that the optimized VIC-machine learning coupled model better simulates in-situ observed streamflow(with an NSE coefficient close to 0.85),solving the overfitting problem of the original model.By simulating streamflow processes at the Yichang Station,located at the headwaters of the Three Gorges Reservoir,we observed an increasing trend in streamflow from 2013 to 2021.Additionally,we used the constructed model for a replay analysis of typical flood events in the Three Gorges Reservoir in 2024(Daning River 2024 No.1 Flood)and further verified the model’s effectiveness in reproducing flood events.
作者
孙祎欣
张强
宋文龙
刘达
SUN Yixin;ZHANG Qiang;SONG Wenlong;LIU Da(Faculty of Geographical Sciences,Beijing Normal University,100875,Beijing;Advanced Interdisciplinary Institute of Environment and Ecology,Beijing Normal University,519085,Zhuhai,Guangdong;China Institute of Water Resources and Hydropower Research,100048,Beijing;Guangdong Institute of Water Resources and Hydropower Research;Guangdong Key Laboratory of Applied Research on Hydrodynamics,510635,Guangzhou,Guangdong,PRC)
出处
《曲阜师范大学学报(自然科学版)》
2025年第4期1-8,共8页
Journal of Qufu Normal University(Natural Science)
基金
三峡后续工作项目(JZ0161A012023)。
关键词
洪水模拟
机器学习
径流过程
洪水过程
三峡库区
flood modeling
machine learning
runoff process
flooding processes
Three Gorges Reservoir Region
