基于自适应神经模糊推理系统和简单水质指标预测供水系统三卤甲烷的浓度  

Prediction of trihalomethane levels in tap water based on adaptive network-based fuzzy inference system and simple water quality parameters

作　　者：洪华嫦[1] 陈敏杰 康家馨 徐昊天 林红军[1] 徐泽琼 孙洪杰[1] 周小玲[1] HONG Huachang;CHEN Minjie;KANG Jiaxin;XU Haotian;LIN Hongjun;XU Zeqiong;SUN Hongjie;ZHOU Xiaoling(College of Geography and Environmental Sciences,Zhejiang Normal University,Jinhua 321004)

机构地区：[1]浙江师范大学地理与环境科学学院,金华321004

出　　处：《环境科学学报》2023年第6期290-299,共10页Acta Scientiae Circumstantiae

基　　金：国家自然科学基金(No.22076171);浙江省基础公益研究计划项目(No.LGF21B070004);浙江师范大学流域地表过程与生态安全重点实验室开放课题(No.KF-2022-05)。

摘　　要：消毒副产物(DBPs)因其潜在的致癌、生殖毒性成为饮用水安全的重要隐患.及时知晓供水系统中DBPs浓度是防控DBPs健康风险的前提.然而目前DBPs仪器监测存在过程繁琐、成本高昂、时间滞后等问题.利用常规水质指标建立高质量的预测模型是重要出路.基于此,本研究选择3个简单的水质指标pH、温度、UVA_(254),利用自适应模糊神经网络(ANFIS)建立了供水系统中三卤甲烷(THMs,最常见的DBPs)分布的预测模型.结果显示,当隶属函数(MF)分别为“gaussmf”、“gbellmf”、“trapmf”,MF数量分别为3、3、3时,建立的总三卤甲烷(T-THMs)、三氯甲烷(TCM)、一溴二氯甲烷(BDCM)模型的预测效果最好.其中,T-THMs、TCM模型质量较高:预测值与实测值的相关系数(r)为0.871~0.880,预测误差(MARE)为11%,预测准确率(N_(E<25%))为92%~95%;与T-THMs、TCM相比,BDCM模型的预测效果略差(r=0.775,MARE=16%,N_(E<25%)=83%),但也在可接受范围.与先前建立的径向基人工神经网络模型相比,用ANFIS方法建立的系列THMs模型预测误差更小、预测准确度更高.本研究将为DBPs的应用型预测模型的建立提供重要参考.Disinfection byproducts(DBPs)have become a significant concern for drinking water safety due to their potential carcinogenic and reproductive toxicity.Timely knowledge of the concentration of DBPs in the water supply system is a prerequisite for controlling the health risks of DBPs.However,the current DBPs monitoring using instrument suffers from problems such as complicated procedures,high costs,and time lag.Developing high-quality predictive models based on conventional water quality indicators is an important solution.In this study,three simple water quality parameters,namely pH,temperature,and UVA_(254),were selected,and an adaptive neuro-fuzzy inference system(ANFIS)was used to establish a predictive model for the distribution of trihalomethanes(THMs,the most common DBPs)in the water supply system.The results show that the best predictive performance of the total trihalomethanes(T-THMs),trichloromethane(TCM),and dibromochloromethane(BDCM)models were achieved when the membership function(MF)was set as"gaussmf","gbellmf",and"trapmf",respectively,and the number of fuzzy subsets for each variable was set as 3.Notably,the T-THMs and TCM models exhibited higher quality,with a correlation coefficient(r)between predicted and actual values of 0.871~0.880,a mean absolute relative error(MARE)of 11%,and a prediction accuracy(N_(E<25%))of 92%~95%.Compared with the T-THMs and TCM models,the predictive performance of the BDCM model is slightly inferior(r=0.775,MARE=16%,N_(E<25%)=83%),but still falls within an acceptable range.Moreover,the THMs models established using the ANFIS method exhibited smaller prediction errors and higher accuracy than the previously established radial basis function artificial neural network models.This study will provide significant reference for the establishment of practical predictive models for DBPs.

关 键 词：供水系统 消毒副产物(DBPs) 三卤甲烷(THMs) 预测模型 自适应神经模糊推理系统(ANFIS) 水质指标 

分 类 号：X32[环境科学与工程—环境工程] X832