Source Identification And Uncertainty Analysis Of DNAPLs-contaminated Groudwater

Groundwater contamination has the characteristic of concealment,and the discovery of groundwater contamination is usually hysteretic,which results in minimal knowledge about groundwater contamination sources.This causes great difficulties to groundwater contamination responsibility confirmation,contamination risk estimation,and remediation strategy design.Therefore,groundwater contamination source identification(GCSI)is especially important.GCSI is accomplished by inversely solving a simulation model that describes contaminant transport in the aquifer based on limited groundwater contamination monitoring data,and finally the information about groundwater contamination sources is obtained,including their number,location,and release history.Here,release history refers to the variation process of release strength of contaminants.GCSI belongs to an inverse problem of mathematical equations in mathematics.Inverse problems are usually nonlinear and ill-posed,because the known information is far less than the unknown information.This is the main difficulty of inverse problems.In this dissertation,the simulation-optimization inverse theories and methods were applied to solving GCSI problem of dense non-aqueous phase liquids(DNAPLs)contamination.DNAPLs,with densities greater than water,have the characteristics of low solubility,high toxicity and interfacial tension.There are many difficulties in DNAPL-contaminated aquifer remediation,such as low contaminant removal rate,long remediation duration,and high remediation cost.Thus,selecting a reasonable and efficient remediation strategy based on the information of DNAPLs source in the aquifer is critical.However,the research on GCSI is still in the development stage,and the research about GCSI for DNAPLs-contaminated site is really rare.Therefore,the further study of unsolved problems in the research fronts of DNAPLs-contaminated aquifer source identification is of important theoretical significance and practical application value.In this dissertation,a hypothetical case and an actual case were combined for in-depth study to explore the feasibility and effectiveness of surrogate-based simulation-optimization method,hybrid homotopy-differential-evolution algorithm,adaptive cyclic update method,and uncertainty analysis in GSCI problem at DNAPLs-contaminated site.Firstly,a DNAPLs-contaminated aquifer multi-phase flow simulation model was preliminarily built,and sensitivity analysis was conducted based on the simulation model to select sensitive aquifer parameters as the variables to be identified.The Latin hypercube sampling method was applied to sampling in the feasible region of the variables to be identified,including contamination source related variables and sensitive aquifer parameters,and the input-output sample dataset was obtained by running the multi-phase flow simulation model.Secondly,according to the input-output sample dataset,support vector regression(SVR),Kriging,and kernel extreme learning machine(KELM)method were applied to building the surrogate model of multi-phase flow simulation model.Differential evolution(DE)algorithm was then applied to determine the optimal weights of surrogate models for building the DE ensemble surrogate model.Thirdly,a nonlinear programming optimization model was built.The DE ensemble surrogate model was embedded in the optimization model as the constraint condition for replacing the input-output relationship of the simulation model.Hybrid homotopy-DE algorithm was applied to solving the optimization model for the identification of contamination source and aquifer parameters.The improved adaptive sampling method was applied to updating the training samples of surrogate model based on the current optimal solution of the optimization model,and then the surrogate model and the optimization model were updated.An adaptive update cycle was built to make the results of sampling,the results of surrogate model and the results of optimization model improved gradually.Finally,aimed at the uncertainty of groundwater quality dynamic monitoring data,the uncertainty analysis of GCSI was conducted by using Monte Carlo method.The probability distribution characteristics of variables related to contamination source,such as the source position and the release history,and aquifer parameters were then obtained for providing more abundant,comprehensive and reliable reference to decision makers.General conclusions drawn from above research are as follows:(1)Kriging surrogate model and KELM surrogate model are more approximate to the multi-phase flow simulation model than SVR surrogate model,and Kriging surrogate model is slightly better than KELM surrogate model.The accuracy evaluation indices of DE ensemble surrogate model are obviously better than three stand-alone surrogate models.The DE ensemble surrogate model is the closest approximation to the simulation model.(2)In comparison with DE algorithm,hybrid homotopy-DE algorithm is better in searching the global optimal solution of the nonlinear programming optimization model,and avoiding the optimization process trapping into premature convergence.The identification accuracy of the contamination source and aquifer parameters is significantly improved by the application of hybrid method.The hybrid of homotopy algorithm and DE algorithm is an effective improvement of the traditional heuristic algorithm.(3)The results of GSCI are significantly improved after the application of adaptive cyclic update method.At the end of the adaptive cyclic update process of hypothetical case,the average relative error of the final identification results is only 2.14%,which is significantly lower than that of the initial identification results.The average relative error of the initial identification results is 3.28%.(4)The methodology applied in this dissertation is effective for accomplishing simultaneous identification of groundwater contamination source and aquifer parameters with small computational load,meanwhile maintaining high computational accuracy.The identified contamination source location provides reliable basis for contamination responsibility confirmation.A multi-phase flow simulation model was built according to the identification results of contamination source and aquifer parameters,and the distribution of contaminants in the aquifer was then calculated,which provides reliable basis for groundwater contamination remediation strategy design and contamination risk estimation.(5)Using deterministic approach can only obtain one set of GCSI results.As the error of groundwater water quality monitoring data is inevitable,which brings great uncertainty to the groundwater water quality monitoring data,the identification results are also with great uncertainty.Through the uncertainty analysis,the probability distribution characteristics of identification results and the confidence interval under different confidence level of identification results were obtained,which provide more abundant,comprehensive and reliable reference to the decision makers.