基于极限学习机算法对太湖东部湖区水源地总磷数据的自动监测分析  

Analysis of automatic monitoring data of total phosphorus in drinking water source in east Taihu Lake based on improved extreme learning machine algorithm

作　　者：崔嘉宇 郭蓉[1] 张悦[1,2] 徐亮 钟声[1] 董圆媛[1] 李心雨[3] CUI Jiayu;GUO Rong;ZHANG Yue;XU Liang;ZHONG Sheng;DONG Yuanyuan;LI Xinyu(Jiangsu Environmental Monitoring Center,Nanjing 210019,China;Jiangsu Suli Environmental Science and Technology Co.,Ltd,Nanjing 210036,China;School of the Environment,Nanjing University,Nanjing 210093,China)

机构地区：[1]江苏省环境监测中心,南京210019 [2]江苏省苏力环境科技有限责任公司,南京210036 [3]南京大学环境学院,南京210093

出　　处：《环境工程学报》2021年第6期2165-2173,共9页Chinese Journal of Environmental Engineering

基　　金：国家水体污染控制与治理科技重大专项(2017ZX07302-003);2018年江苏省环保科研课题(重大技术攻关类)(2018008);江苏省环境监测科研基金项目(2003)。

摘　　要：对太湖东部湖区金墅港和渔洋山水源地总磷自动监测值与实验室监测值进行比对分析,基于相关性分析,选取对自动监测总磷浓度影响较大的浊度、蓝绿藻和叶绿素等指标,提出可选择极限学习机隐层节点的改进极限学习机模型(IELM)修正自动监测总磷数据。结果表明,总磷自动监测值相对实验室监测值偏差较大,绝对误差范围在0.05~0.112 mg·L^(-1),平均绝对误差为0.017 mg·L^(-1)。若以相对误差小于或等于30%作为比对合格的标准值,总磷自动监测数据合格率仅为52.9%。经过IELM算法模型训练和测试,训练均方误差为0.0000735,测试均方误差为0.000103。经过模型修正后的自动监测总磷数据更接近实验室监测值,其平均绝对误差降低0.026 mg·L^(-1),平均相对误差降低45%。按比对误差30%计算,模型修正后的总磷自动监测数据合格率为92.0%,可应用于自动监测总磷数据的修正。A comparative analysis was performed on the concentration of total phosphorus measured in the Jinshugang and Yuyangshan drinking water source areas in East Taihu Lake by automatic water monitoring and the laboratory method.Three key factors were found to influence the automatic monitoring of total phosphorus according to a correlation analysis,i.e.water turbidity,algal density and chlorophyll-a.The three parameters were introduced into the improved extreme learning machine model(IELM)for correction of automatic water monitoring data of total phosphorus.Compared with laboratory method,the total phosphorus concentrations measured by automatic water monitoring were shown to have relatively large errors with absolute error ranging between 0.05 mg·L^(-1) and 0.112 mg·L^(-1) and the mean absolute error being 0.017 mg·L^(-1).The training error was0.0000735 and test error was 0.000103 after the training and testing of the IELM model.The measurement results by the application of IELM showed better performance.With 30% of relative error rate as judgment criteria,the eligible rates increased from 52.9% to 92.0%,the absolute error decreased by 0.026 mg/L on average,and the relative error rate dropped by 45%.The results in this study show the promise of application of IELM model in correction of automatic monitoring of total phosphorus in the field.

关 键 词：总磷 自动监测 改进极限学习机 隐层节点 修正数据 

分 类 号：X832[环境科学与工程—环境工程] X524