Prediction Of Water Quality In MBR Treatment Of High Salt Ammonia Nitrogen Wastewater Based On Machine Learning

Intelligent sewage treatment system mainly includes intelligent monitoring,intelligent control and intelligent management.In terms of intelligent monitoring,real-time monitoring and data analysis of water quality,flow,temperature,PH value and other key parameters in the sewage treatment process are realized through sensors,data acquisition and remote monitoring technology.In terms of intelligent control,automatic control technology is adopted to realize automatic control of all links in the sewage treatment process.In terms of intelligent management,cloud computing,big data analysis,artificial intelligence and other technologies are used to analyze and process the data in the sewage treatment process to improve management efficiency and decision-making ability.At present,China’s sewage treatment has achieved online monitoring and automatic control of sewage.However,enterprises still need to spend a lot of manpower and financial resources on the causes of water quality changes and the countermeasures to be taken.There are many problems such as high time cost and delayed decision-making.And machine learning is an artificial intelligence technology,which enables computers to learn from data and improve performance to achieve independent decision-making and prediction.Therefore,on the basis of the research on the characteristics of bacteria and process parameters of high-salt ammonia nitrogen wastewater by the research group,this paper uses machine learning to predict a number of water quality indicators.Provide a real-time monitoring strategy based on big data for enterprise and factory wastewater treatment.The main contents are as follows:(1)In this paper,17 parameters of reactor water quality index(including DO,HRT,pH,water temperature,COD and treatment efficiency of inlet and outlet water,ammonia nitrogen of inlet and outlet water,nitrite nitrogen of inlet and outlet water,nitrate nitrogen of inlet and outlet water,total nitrogen of inlet and outlet water and treatment efficiency,C/N),data source,incubation days and salinity are taken as the research object,and the data recorded in five batches of reactors are sorted and pretreated,Select five kinds of regression algorithms(linear model,ensemble learning,nearest neighbor model,support vector machine and kernel ridge regression)that are more common in five kinds of machine learning to build models using python programming language,and analyze the applicability of the algorithm through learning curve and MSE evaluation score.Eight characteristic variables are determined according to the actual process.(2)The single-target prediction model and the multi-objective prediction model are established based on whether the effluent ammonia nitrogen is taken as the characteristic variable.Select algorithms that perform well in the exploration phase(RF,GB,KNN,SVR,KRR),use a single parameter adjustment curve to determine the range of built-in adjustable parameters for the algorithm,and use the grid search method to find the optimal built-in parameters within this range.The results show that the super-parameter optimization efficiency of the multi-objective prediction model is higher,and the algorithm that performs relatively well in target prediction after parameter optimization is GB.The MSE of effluent ammonia nitrogen,total nitrogen,nitrate nitrogen,nitrite nitrogen and effluent COD are 24.2,110.9,0,115.8 and 21.5 respectively.(3)The feature importance of GB algorithm is analyzed,and the influence of feature variables on target prediction value is compared numerically.In the single-target analysis,the characteristic variables that have a greater impact on the ammonia nitrogen in the effluent are pH,ammonia nitrogen in the influent and COD in the influent,accounting for 25%,17.5% and7.5% respectively;In the multi-objective analysis,the characteristic variables that have a greater impact on total nitrogen and effluent nitrite are temperature,salinity and effluent ammonia nitrogen,accounting for 34.0% and 44.3%,18.9% and 19.2%,16.7% and 17.5%respectively;Salinity,ammonia nitrogen and temperature are the characteristic variables that have a greater impact on nitrate nitrogen,accounting for 68.3%,9.6% and 8.5% respectively;.The characteristic variables that have a greater impact on the COD of effluent are DO,COD of influent and ammonia nitrogen of effluent,accounting for 27.6%,21.4% and 19.0% respectively.(4)This paper proposes a data set contribution analysis method(DCD)based on the combination of machine learning and evaluation indicators.This method takes into account the impact of each data set on the prediction target,gives corresponding weights,and provides a reasonable and interpretable way to combine data from different sources.According to the analysis of the contribution of each sub-data set to the prediction results by DCD evaluation,the DCD of nitrite nitrogen in source "1","2" and "3" is 0.86,0.5 and 0.80 respectively,the DCD of total nitrogen in source "1" and "3" is 0.78 and 0.52 respectively,and the DCD of effluent COD in source "1" and "4" is 0.022 and 0.021 respectively.The results show that the salinity impact of sample data is helpful for the model to predict the change of saline water quality.