水资源与水工程学报2026，Vol.37Issue(4)：29-38,10.DOI:10.11705/j.issn.1672-643X.2026.04.04
基于可解释机器学习的用水量驱动因素分析
Analyzing driving factors of water use based on interpretable machine learning methods:
摘要
Abstract
The rapid development of economy and society leads to intensified water supply-demand con-flicts,which threatens the security and sustainable development of regional water resources.Revealing the driving mechanisms of regional water use is crucial for achieving differentiated water resources man-agement and efficient allocation.This study focuses on the water use status in 12 prefecture-level cities in Inner Mongolia Autonomous Region based on socio-economic and water use data from 2002 to 2023.Six machine learning models,such as LightGBM,CatBoost,and Random Forest,were constructed and their predictive performances compared.SHAP and ALE,two interpretable machine learning methods,were employed to identify the key drivers of water use and their nonlinear interaction effects.Furthermore,hi-erarchical clustering was used to uncover the spatial heterogeneity and coupling patterns of water use driv-ing mechanisms.The results show that the LightGBM model produces the highest accuracy,with an R2 of 0.989.The driving mechanisms of water use in different cities exhibit significant spatial heterogeneity.Industrialized urban clusters(Hohhot,Baotou,Ordos)are mainly driven by economic scale and urbani-zation;agricultural and pastoral regions(Chifeng,Bayannur,Ulanqab,etc.)are primarily influenced by GDP,effective irrigated area,and livestock scale;and arid zones(Alxa and Wuhai)are sensitive to water use per unit GDP and population density.In addition,multiple driving factors exhibit nonlinear in-teractions between each other,such as the synergistic effects between urbanization and water use efficien-cy,and between irrigation and livestock expansion.These findings can provide theoretical support and quantitative evidence for formulating region-specific water-saving policies and optimizing water resources allocation.关键词
用水量/可解释机器学习/SHAP/ALE/驱动机制/内蒙古Key words
water use/interpretable machine learning/SHAP/ALE/driving mechanism/Inner Mongo-lia Autonomous Region分类
建筑与水利引用本文复制引用
杨柳悦,张凯,粟晓玲,褚江东,崔德新,贾玉博..基于可解释机器学习的用水量驱动因素分析[J].水资源与水工程学报,2026,37(4):29-38,10.基金项目
内蒙古自治区水利科项目(202501010505A) （202501010505A）
