基于机器学习的长江流域城市初期雨水径流污染特征及预测研究  

Characteristics and predictions of initial rainwater runoff pollution in the Yangtze River Basin based on machine learning

作　　者：陈亚松 侯兴[2,3] 赵云鹏 张文龙[2] 朱雅婷 高煜 李轶[2] CHEN Yasong;HOU Xing;ZHAO Yunpeng;ZHANG Wenlong;ZHU Yating;GAO Yu;LI Yi(Yangtze River Eco-Environmental Engineering Research Center,China Three Gorges,Wuhan 430010,China;College of Environment,Hohai University,Nanjing 210098,China;Institute of Water Science and Technology,Hohai University,Nanjing 210098,China)

机构地区：[1]中国长江三峡集团有限公司长江经济带生态环境国家工程研究中心,武汉430010 [2]河海大学环境学院,南京210098 [3]河海大学水科学研究院,南京210098

出　　处：《环境工程》2025年第2期62-73,共12页Environmental Engineering

基　　金：中国长江三峡集团有限公司自主科研项目“长江流域典型城市初期雨水与溢流污染的综合治理技术研究”(WWKY-2021-0475)。

摘　　要：根据地域特征与降水特征预测初期雨水径流污染对海绵城市建设具有一定意义。收集了长时间、大尺度的城市特征、降水特征、下垫面性质与初期雨水水质数据,将13个长江流域典型城市按相似与相异度聚类,并采用主坐标分析验证了结果的合理性;探究了不同类型城市初期雨水径流污染的统计学特征,并解析了驱动因子;最终建立了初期雨水径流污染预测的随机森林回归模型,并采用南京市实测数据进行验证。结果表明:长江流域城市根据地形、气候、经济和初雨污染等特征可分为3种类型;年降水量、单次降水量、平均雨强、雨前干期和下垫面不透水性是影响初期雨水径流污染物浓度的主要因素;随机森林回归模型的拟合R^(2)>0.68,且南京市实测值与随机森林预测结果的误差均不超过15%,说明预测结果较为准确,可用于长江流域不同类型城市的初期雨水地表径流污染预测。It is significant for sponge city construction to predict the pollution of initial rainwater runoff in line with regional and precipitation characteristics.In this paper,a vast amount of extensive and long-term data concerning various aspects such as urban characteristics,precipitation features,the nature and properties of underlying surfaces,and the quality of initial rainwater were meticulously collected.Thirteen typical cities located in the Yangtze River Basin were clustered and grouped based on their complex degrees of similarity and dissimilarity.The rationality and validity of the clustering results were carefully verified and authenticated through principal coordinate analysis.Subsequently,the statistical characteristics of the pollution of initial rainwater runoff in different types of cities were thoroughly investigated and explored,and the underlying and influential driving factors were dissected and analyzed in detail.This involved not only a comprehensive examination of the correlation between different variables but also an exploration of the potential mechanisms.Eventually,a highly sophisticated random forest regression model specifically designed for predicting the pollution of initial rainwater runoff was established and then validated and verified with the measured data obtained from Nanjing.The results and findings indicated that the cities in the Yangtze River Basin can be effectively categorized into three distinct types according to terrain,climate,economy and the specific situation of initial rainwater pollution.Annual precipitation,single precipitation,average rainfall intensity,dry period duration before rainfall and impermeability degree of underlying surfaces are the principal factors on pollutants concentration in initial rainwater runoff.The fitting R-squared values of the random forest regression model were all greater than 0.68,and the errors between the measured values in Nanjing and the predicted results of the random forest were all within 15%,which convincingly demonstrated that

关 键 词：长江流域 初期雨水 机器学习 聚类分析 随机森林回归 

分 类 号：X522[环境科学与工程—环境工程]