Inverse Identification Of LNAPLs Contamination Source In Groundwater

Petroleum and its products often leak into the aquifer arising in improper handling or unexpected accidents,which will cause contamination to groundwater.Unlike surface water contamination,groundwater contamination is hidden deeply below the surface of the land,and is stored and migrate in the pore medium of rock and soil.Because of its the concealment of occurrence and lag of discovery,groundwater contamination is difficult to be detected in time when contamination occurs,which leads to the aquifer contamination source information is not known and mastered.This brings great difficulties to the identification of the perpetrators of groundwater contamination,contamination risk assessment,contaminant migration prediction,and contamination remediation plan design.Therefore,it is very important to carry out the research on the inverse identification of groundwater contamination.The inverse identification of groundwater contamination source refers to use of limited field monitoring data(water level and concentration)to solve the numerical simulation model describing groundwater contamination in reverse,based on the auxiliary work of data collection,field investigation and qualitative analysis,to determine the information of contamination source in aquifers,including the number,location and discharge history of contamination source.The inverse identification of groundwater contamination belongs to the inverse problem of mathematical equation,which often has the characteristics of ill posed and nonlinear.At present,the inverse identification of groundwater contamination is still in the development stage,and there are few reports on the inverse identification of groundwater light non-aqueous phase liquids(LNAPLs).Most of LNAPLs have the characteristics of low water solubility,high toxicity,less specific gravity than water,easy to volatilize and diffuse,and easy to be degraded by microorganisms.After get into the groundwater,it will cause water to become unsafety and destroying ecological environment.Therefore,it has significant importance to develop a reasonable and efficient remediation scheme for LNAPLs contamination.However,the prerequisite of identifying and mastering the information of LNAPLs contamination source in aquifers is an important step for formulating contamination remediation plans.Therefore,it is of great theoretical significance and practical application prospect to study the source identification of LNAPLs contamination in groundwater.This paper adopts a research method that combines theoretical analysis and practical examples,through the comprehensive use of theories and methods such as simulation-optimization method,optimal complementary noise reduction method,artificial intelligence set pair surrogate model,adaptive hybrid gray wolf optimization algorithm,and Monte Carlo method,to conduct a deep research on the scientific problems that still need to be solved in the research frontier of the inverse identification groundwater LNAPLs,expand and enrich the theories,methods and technical connotations of the inverse identification groundwater contamination.Firstly,based on the auxiliary work of data collection,field investigation and qualitative analysis,the geological and hydrogeological conditions of the study area are generalized,and the conceptual model of the study area is established based the specific conditions of the study area.By making full use of the previous work results and making preliminary estimates of the aquifer parameters and contamination source information that need to be determined,and the groundwater LNAPLs contamination multiphase flow numerical simulation model describing the contaminant transport is preliminarily established.After that,in order to improve the noise reduction effect of dynamic monitoring data,the optimal complementary noise reduction method based on empirical mode decomposition method,ensemble empirical mode decomposition method and complementary ensemble empirical mode decomposition method is established in this paper,and then applied to the noise reduction experiment of hypothetical example dynamic monitoring data,analyzing its applicability and effectiveness,and then applied it to the actual example.The noise reduction of dynamic monitoring data lays a solid foundation for the follow-up research.Then,the sensitivity analysis method is used to select the simulation model parameters which have great influence on the output of the multiphase flow numerical simulation model.The selected simulation model parameters and groundwater contamination source information can be obtained by solving it as the variables,and samples them within its value range through Latin hypercube sampling method.In order to obtain training samples and test samples,the samples are put into the multiphase flow numerical simulation model in turn,and perform the forward calculation.The training samples and test samples are used to train the short-term memory neural network surrogate model and the deep belief neural network surrogate model respectively,on which the artificial intelligence set pair replacement model is established.The accuracy and applicability of artificial intelligence set pair surrogate model are analyzed by comparing artificial intelligence set pair alternative model with other artificial intelligence models based on single method.Finally,the nonlinear programming optimization model is developed,and the artificial intelligence set pair substitution model is integrated into the optimization model as an equality constraint,and then explores the effective solution of nonlinear programming optimization model.Levy flight random walk strategy,metropolis acceptance criterion and adaptive weight strategy are introduced to improve the traditional gray wolf optimization algorithm.The adaptive hybrid grey wolf optimization algorithm based on Levy flight and metropolis acceptance criteria are applied to solve the optimization model,so as to obtain simulation model parameters and contamination source information of the identification results.At the same time,the applicability of adaptive hybrid gray wolf optimization algorithm is analyzed.In addition,the parameters in the simulation model are randomly perturbed according to the analysis results of the simulation model parameters.The observe to random change of parameters that resulting in uncertainty of result is analyzed through the combination of Monte Carlo method and simulation-optimization method.The digital characteristics,probability distribution and confidence intervals of contamination source information under different confidence levels are obtained,which can provide more abundant reference for decision makers.Based on the above research contents,the following main conclusions are drawn:(1)In order to improve the denoising effect of dynamic monitoring data,the optimal complementary denoising method is constructed on the basis of empirical mode decomposition,ensemble empirical mode decomposition and complementary ensemble empirical mode decomposition.The noise reduction effect of the optimal complementary noise reduction method is better than that of the three single methods,which is more suitable for the noise reduction of groundwater dynamic monitoring data.(2)The long-term memory neural network surrogate model and deep belief neural network surrogate model have higher approximation accuracy than the extreme learning machine surrogate model and Kriging surrogate model for multiphase flow simulation models.Base on the long-term memory neural network surrogate model and deep belief neural network surrogate model,the artificial intelligence set pair surrogate model is established.The simulation results show that the approximation accuracy of artificial intelligence set pair alternative model is better than other four single artificial intelligence models.The artificial intelligence set pair surrogate model has better fitting ability for multiphase flow numerical simulation model with many kinds of variables and complex nonlinear mapping relationship between input and output.(3)The effective solution of nonlinear programming optimization model is explored.The Levy flight random walk strategy,metropolis acceptance criteria and adaptive weight strategy are applied to the traditional gray wolf optimization algorithm to improve the traditional gray wolf optimization algorithm.The adaptive hybrid gray wolf optimization algorithm based on Levy flight and metropolis acceptance criteria can quickly search for the global optimal solution without falling into the local optimal solution,which improve the accuracy of groundwater LNAPLs contamination identified results.(4)Based on the simulation optimization method,only the unique identified result can be obtained.In order to analyze the influence of random variation of parameters on the uncertainty of contamination source identified results,Monte Carlo method and simulation optimization method are combined to analyze the uncertainty of LNAPLs identification results.Numerical characteristics,probability distributions and confidence intervals of contamination source information under different confidence levels can be obtained.in order to provide richer references for decision makers.