随机森林回归模型在地下水水质评价的新应用  

New Application of Random Forest Regression Model on Groundwater Quality Assessment

作　　者：许飞青 李潇 李凯 于喆 郭亚杉 赵志强 XU Feiqing;LI Xiao;LI Kai;YU Zhe;GUO Yashan;ZHAO Zhiqiang(Beijing Institute of Geology and Mineral Exploration Information Center,Beijing 100195;Beijing Key Laboratory of Urban Spatial Information Engineering,Beijing 100038;China Aerospace Academy of Architectural Design ResearchCo.,Ltd.,Beijing 100162)

机构地区：[1]北京市地质矿产勘查院信息中心,北京100195 [2]城市空间信息工程北京市重点实验室,北京100038 [3]航天建筑设计研究院有限公司,北京100162

出　　处：《地质与勘探》2023年第2期408-417,共10页Geology and Exploration

基　　金：城市空间信息工程北京市重点实验室经费项目(编号:20210219)资助。

摘　　要：地下水水质评价是地下水研究的一个重要课题,研究更精确、更适用的地下水水质评价模型具有重要意义。近年来应用较多的评价模型存在适用性小、精确度低的问题,其中主要原因在于现有的水质评价标准中,不同因子的等级分类标准不一,同一因子不同等级取值区间不一致等问题,而现有模型无法克服不同因子自身评价等级不匹配以及无法有效处理多重因子之间的相互影响。针对水质评价存在的问题,本文将研究一个能有效克服上述问题,且精确度较高、适用性更广的水质评价模型。基于北京市地下水监测网中污染源监测网的数据,选取大兴区作为研究区域,以大兴区2018年至2020年持续进行监测的39个点位作为研究点,根据监测点历年监测水样,选取化验数据中影响水质环境较为明显的pH值、Cl^(-)、NO^(-)_(3)、SO_(4)^(2-)、Na^(+)、NH^(+)_(4)、Mn^(-)、耗氧量(COD Mn)、总硬度、溶解性总固体这10项指标作为水质评价因子,利用随机森林回归算法对其进行评价分析,得到研究区2018~2020年地下水水质等级指数,显示出研究区水质质量总体呈先变差后变好的趋势,其中Na^(+)、SO_(4)^(2-)、NO^(-)_(3)对水质等级影响较大。随机森林与常用的支持向量机(SVM)和人工神经网络算法以及高斯过程方法做对比,随机森林算法数据对水质评价具有更高的分类准确性和更高的灵敏度,适用于地下水水质评价。Groundwater quality evaluation is an important subject in groundwater research.It is significant to study a more accurate and more suitable model for groundwater quality evaluation.In recent years,many evaluation models have problems of low applicability and accuracy.The main reason is that in the existing water quality evaluation standards,the classification standards of different factors are different,and the value ranges of different grades of the same factor are inconsistent.However,the existing models can not overcome the mismatch of evaluation grades of different factors and can not effectively deal with the interaction between multiple factors.In view of the existing problems of water quality evaluation,this paper will study a water quality evaluation model which can effectively overcome the above problems and has higher accuracy and wider applicability.Based on the data of the pollution source monitoring network in the groundwater monitoring network of Beijing,Daxing District was selected as the research area,and 39 continuous monitoring points in Daxing District from 2018 to 2020 were selected as the research sites.According to the water samples in the monitoring points over the years,ten indexes,including pH value,Cl^(-),NO^(-)_(3),SO_(4)^(2-),Na^(+),NH^(+)_(4),Mn^(-),oxygen consumption(COD Mn),total hardness and dissolved total solids,which significantly affect the water quality environment in the laboratory data were selected as the evaluation factors,and were evaluated and analyzed by random forest regression algorithm.The groundwater quality grade index of the study area from 2018 to 2020 was obtained,which shows that the water quality of the study area generally becomes worse and then better,among which Na^(+),SO_(4)^(2-)and NO^(-)_(3)have a great influence on the water quality grade.Compared with the commonly used support vector machine(SVM)and artificial neural network algorithm and gaussian process method,the data of random forest algorithm has higher classification accuracy and sensitivity for

关 键 词：水质评价 地下水 随机森林 神经网络 回归 大兴区 北京 

分 类 号：P641[天文地球—地质矿产勘探] X824[天文地球—地质学]