Identification Of DNAPLs-Contaminated Groundwater Pollution Sources Based On Bayes Theory

The leakage of petroleum products in the process of transportation,unreasonable discharge of industrial wastewater and the sudden accidents in the process of industrial production may lead to pollutants entering the groundwater environment,endangering the safety of drinking water and ecological environment.Different from surface water pollution or air pollution,groundwater pollution has the characteristics of concealment and lag in discovery.Timely understanding and mastering the relevant conditions of groundwater pollution sources,such as the number of pollution sources,space location and release history,are the important premise for the remediation and treatment of groundwater pollution,risk assessment and responsibility identification.Therefore,the study on the inversion identification of groundwater pollution sources is very important.Identification of groundwater pollution sources refers to the inversion of the mathematical simulation model describing groundwater pollution through the limited monitoring data(water level and concentration),as well as auxiliary information such as investigation and professional knowledge.So information such as the number of pollution sources,spatial location and release history can be identified.Identification of groundwater pollution sources is a typical inversion problem.Most of the inversion problems are nonlinear and ill posed,which makes it difficult to solve.Therefore,inversion and identification of groundwater pollution sources is still a challenging research topic in groundwater pollution study.Based on this,to meet the actual needs of groundwater pollution remediation,this paper combines theoretical analysis and practical example to carry out research on the scientific issues to be solved in the frontier research of dentification of dense non-aqueous phase liquid(DNAPLs)of petroleum organic pollutants.Firstly,through literature research and analysis,a research method system of identification of DNAPLs-contaminated groundwater pollution sources based on Bayesian theory was constructed.The research method system consisted of three parts:(1)The optimal experimental design method based on Bayesian theory was constructed,which was applied to determine the optimal monitoring well location in the process of pollution sources identification,so as to obtain the monitoring data with the strongest correlation with pollution sources identification.(2)The DREAM_Galgorithm,which was an improved algorithm of DREAM algorithm that was a multi chain Markov chain Monte Carlo(MCMC)algorithm,was constructed and applied for pollution sources identification,so as to accelerate the convergence speed of sampling process and improve the accuracy of identification.(3)The deep residual network(Res Net)was applied to establish the surrogate model for multiphase flow numerical simulation model,and compared with the surrogate models established by Kriging(KRG)method and support vector regression(SVR)method to analyze the applicability of Res Net surrogate model.The purpose of surrogate modeling was to reduce the huge computational load caused by running simulation model for thousands of times in the process of optimal experimental design and identification.Then,ideal numerical examples with different situations were designed to conduct theoretical analysis and research on the applicability and effectiveness of the research method system through the example test.Finally,on the basis of the theoretical analysis and research results,an actual pollution site in Northwest China was taken as the practical example.And the identification of DNAPLs-contaminated groundwater pollution sources in the study area were systematically studied combined with field investigation and sampling monitoring according to the actual needs of the study area,so as to identify the unknown pollution source information and aquifer parameters.Through the above research,this paper draws the following conclusions:(1)Whether the study on ideal numerical example or practical example,the precision of Res Net surrogate model was higher than that of KRG surrogate model and SVR surrogate model,which showed that the approximation accuracy of Res Net surrogate model to the simulation model was higher.The problem of complex nonlinear mapping between input and output of the simulation model can be effectively solved by introducing the deep learning theory and method.(2)The optimal experimental design method proposed in this paper was reasonable and effective.The obtained optimal monitoring scheme could effectively improve the accuracy of pollution sources identification by applying Bayesian test design,relative entropy,0-1 integer programming optimization model and other theories and methods for optimal design of monitoring well location in the process of pollution sources identification.(3)By combining the GLUE method and DREAM algorithm,the DREAM_Galgorithm was constructed,and the applicability and effectiveness of the algorithm were analyzed through the ideal numerical examples.The results showed that:compared with DREAM algorithm,DREAM_Galgorithm could not only effectively accelerate the convergence speed of sampling process,but also effectively improve the identification accuracy.DREAM_Galgorithm effectively solves the problem that DREAM algorithm do not optimize the initial population of iteration.(4)The constructed DREAM_G-MCMC method was used to identify the pollution sources for the case study area,and the accuracy of the obtained identification results was reliable.This showed that the identification results could further improve the mathematical simulation model of the study area,making it more practical.To sum up,this paper not only enriches and expands the theoretical basis and technical connotation of DNAPLs-contaminated groundwater pollution sources identification,but also provides a scientific basis for the monitoring and remediation of the contaminated site.