基于堆叠机器学习的非充分数据条件下流域径流还原方法研究:以沮漳河为例  

A watershed runoff reconstruction method based on stacked ensemble machine learning under incomplete data:A case study of Juzhanghe River

作　　者：王楠 李明蔚 陈首志 宋儒霖 张璇 郝芳华 付永硕 WANG Nan;LI Mingwei;CHEN Shouzhi;SONG Rulin;ZHANG Xuan;HAO Fanghua;FU Yongshuo(College of Water Sciences,Beijing Normal University,Beijing 100875,China;Hydropower Planning and Design Institute Co.,Ltd.,Beijing 100120,China;College of Urban and Environmental Sciences,Central China Normal University,Wuhan 430079,Hubei,China)

机构地区：[1]北京师范大学水科学研究院,北京100875 [2]水电水利规划设计总院有限公司,北京100120 [3]华中师范大学城市与环境科学学院,湖北武汉430079

出　　处：《水利水电技术（中英文）》2024年第10期85-97,共13页Water Resources and Hydropower Engineering

基　　金：国家自然科学基金区域创新发展联合基金(U21A2039);国家自然科学基金重点项目(42330515)。

摘　　要：【目的】为实现数据非充分条件下的径流还原,【方法】提出一种在用水量数据不完全时基于水量平衡原理构建堆叠机器学习模型计算河川断面的天然径流的方法,并以沮漳河为例计算了河溶水文站断面的天然径流。首先选取与农业、工业、生活用水消耗量具有相关性的指标构建了特征变量指标体系,与研究时段内有缺失的用水量数据一同输入堆叠机器学习模型中,获得连续的用水量数据。再基于水量平衡原理,在水文站实测径流的基础上加减由人类活动引起的径流变化量,计算得到断面的天然径流。【结果】堆叠机器学习模型预测缺失的用水量分量的相对误差分别为0.62%、0.03%、4.95%,经还原计算后河溶站断面在2002—2020年的平均天然径流量为54.5 m3/s,天然径流深为501.3 mm。【结论】提出的方法可实现取用水量数据在时空尺度上有缺失地区的天然径流还原,对区域水资源综合管理和优化配置具有重要意义。[Objective]To achieve runoff reconstruction under incomplete data,[Methods]a method is introduced to fill spatio-temporal gaps in water consumption data.Drawing on the water balance laws,a stacked machine learning model is created.The model is then applied to calculate the natural runoff of the Herong hydrological gauging station as a case study.Initially,a set of feature variables correlating with agricultural,industrial,and domestic water consumption is identified to create a comprehensive feature variable indicator system.The system uses intermittently available water consumption data as input into stacked ensemble machine learning model to produce a continuous water consumption dataset.Adhering to the water balance principle,natural runoff is calculated by adjusting measured runoff attributed to anthropogenic activities.[Results]The stacked ensemble machine learning model produced relative errors of 0.62%,0.03%,and 4.95%for agricultural,industrial,and domestic water,respectively.The average annual natural runoff at the Herong hydrological gauging station from 2002 to 2020 was 54.5 m3/s,with a natural runoff depth of 501.3 mm.[Conclusion]The proposed method enables the reconstruction of natural runoff in areas with missing water consumption data across temporal and spatial scales,and is of great significance for the comprehensive management and optimal allocation of regional water resources.

关 键 词：径流 径流还原 水量平衡原理 堆叠机器学习 用水量 人类活动 沮漳河流域 

分 类 号：P333[天文地球—水文科学]