Study On Water Quality Evaluation Of Fuzhou City Based On XGB＿BN Model

In recent years,water pollution has been paid more and more attention by the society.Water quality evaluation,as an indispensable part of water resource environmental management,has been the focus of research.Based on the water quality monitoring data of three sections in Fuzhou from January 2015 to December 2020,this paper mainly uses python3.8 to carry out data mining and establishes water quality evaluation model to explore the water quality status of Fuzhou.In order to select the appropriate index characteristics as the initial node of Bayesian network model,taking the water quality monitoring data of Fuzhou City from 2015 to2019 as training samples and water quality grade Y as the target variable,we choose the mutual information,random forest algorithm and xgboost algorithm are selected to construct the characteristics respectively,and the modeling features were obtained as follows:（1）COD_Mn,Hg,An,Ph,BOD5,DO,COD,F-,Pe,CN,Zn,P;（2）P,COD,N,NH₃-N,Cu,DO,COD_Mn;（3）P,N,Hg,NH₃-N,COD.After getting the initial nodes of the model by different feature extraction methods,the initial models are input into Bayesian model respectively.Based on BIC scoring principle,the optimal network structure is determined by mountain climbing search,and then Bayesian estimation is used to learn network parameters,and the models are established.The models then are applied to the 12 month water quality evaluation of Fuzhou City in 2020,and the results show that the prediction accuracy of the BN model,RF＿BN and XGB＿BN model are 77.78%,88.89%and 91.67%respectively.Therefore,XGB＿BN model has the best prediction effect.According to XGB＿BN modeling principle,the XGB＿BN model is rebuilted for Minjiang,Aojiang and Longjiang river basins with the prediction accuracy of 100%,100%and 91.67%respectively,which implies that XGB＿BN model is universal and has good prediction effect.In addition,the correlation analysis of water quality factors is carried out through the network structure chart.The indicators variables directly related to water quality grade Y in Minjiang river basin are:COD,Hg,P and DO,in addition,COD has direct relationship with Hg and NH₃-N;the indicators in Aojiang river basin that can directly affect the water quality grade are P and COD,in addition,N directly acts on NH₃-N,and has indirect influence on water quality grade;P in Longjiang river basin can directly affect the water quality level of the basin,NH₃-N can directly affect P,and has indirect relationship with water quality grade.The sensitivity analysis of Bayesian network is carried out by using Ge NIe2.0software and the influence of water quality index on water quality grade is studied.The results show that in Minjiang river basin,the influence of water quality index on water quality grade is as follows:Hg and Do are the most influential,NH₃-N follow,P and COD are the least.This shows that heavy metal pollution and industrial pollution are the main causes of serious pollution in Minjiang River Basin;in Aojiang river basin,the influence degree of water quality index on water quality grade is COD>N>P.This shows that the main pollution sources of Aojiang River Basin are:aquaculture pollution,organic pollution;In Longjiang river basin,the influence degree is P>NH₃-N.This shows that the main pollution sources in Longjiang River Basin involve domestic sewage pollution and surface runoff pollution.Finally,taking the water quality of the Heshan ferry section in Aojiang river basin as an example,the prediction accuracy of the COD grade range is 83.33%under the condition of known N.This method is of reference significance to the evaluation,prediction and decision-making of water resources and environment management system in three major basins of Fuzhou.