The Optimization Design And Uncertainty Analysis Of DNAPLs-contaminated Groundwater Remediation Based On The Adaptive Surrogate Model

Because of the exploitation of crude oil,the leakage of gas pipelines,the indiscriminate discharge of oily wastewater,the sudden leakage of petroleum products and other reasons,so that a large amount of petroleum organic pollutants penetrated into the aquifer,which have caused serious groundwater pollution.Most liquid organic contaminants are nearly immiscible in water,which often referred to as non-aqueous phase liquids(NAPLs).DNAPLs is one kind of NAPLs that its density is greater than water(Dense Non-Aqueous Phase Liquids,DNAPLs),there is often a common movement of water phase,oil(DNAPLs)phase,gas phase of DNAPLs contaminated groundwater,which will constitute a multiphase fluid system.Due to the special properties of DNAPLs such as high density,high interfacial tension and low solubility,the remediation difficulty of DNAPLs contaminated groundwater has increased.In recent years,Surfactant Enhanced Aquifer Remediation(SEAR)is developed based on the pump-and-treat techniques.It can effectively improve the DNAPLs remediation efficiency,significantly reduce the DNAPLs treatment time,and it was widely considered as one of the most promising techniques to remediate DNAPL contaminations.DNAPLs-contaminated groundwater remediation not only need to take the technical method for effective removal of contaminants into account,but also need to consider many factors such as the feasibility of the optimal design and economic cost.Optimal management of groundwater contaminant remediation can get an optimal design which achieves the goal of the optimal remediation effects with the costs as low as possible.Therefore,optimal management of groundwater remediation attracts more and more attention.Aiming at the problem of DNAPLs-contaminated groundwater remediation,taking the real DNAPLs-contaminated groundwater remediation as the research case,the optimal management was carried out.In this study,on the basis of a multiphase flow numerical simulation model for surfactant-enhanced DNAPLs contaminated groundwater,adaptive sampling method was used to get the samples for the controllable input variables of the multiphase flow numerical simulation model in the feasible region,and the multiphase flow numerical simulation model was called to prepare for input-output data sets.Adaptive sampling included the initial sampling stage and iterative sampling stage,Latin Hypercube Sampling was used as initial sampling method.And according to initial input-output data sets,polynomial regression,radial basis function neural network,Kriging and support vector regression were respectively used to build the surrogate model of multiphase flow numerical simulation model,and the accuracy of’ four surrogate models were evaluated,the surrogate model with highest accuracy was selected as a surrogate model of multiphase flow numerical simulation model,which would be embedded into the optimization model.Then,considering the uncertainty of surrogate model,the probabilistic constrained programming optimization model of surfactant-enhanced DNAPLs-contaminated groundwater remediation was formulated based on the surrog,ate model.Using simulated annealing algorithm to solve the optimization model,and get the optimal remediation design.Finally,the optimal solution method was used as iterative sampling,the current optimal solution was added to the initial data sets as a new sample,re-training the parameters of surrogate model and getting new optimal solution,continuing the iterative sampling until meet the convergence criteria,the output result was the final DNAPLs-contaminated groundwater remediation optimization design.Main conclusions obtained from the paper are as follows:(1)Adaptive sampling method avoids the problems that the number of sampling is too small to meet the accuracy of surrogate model and the number of sampling too large to prolong the calculating time.At the same time,through the iterative sampling process it focuses on the current optimal solution area to sample,which provide more searching information.In the process of adaptive sampling,as the effect of sampling is improving,surrogate model and optimization model are drived to be continuously improved,thus ensuring the effectiveness of the optimal solution.Compared with the traditional sampling methods,even with the same sampling number,the approximation accuracy of surrogate model based on adaptive sampling method to simulation model is significantly higher.(2)Compared the approximate accuracy with polynomial regression model,RBFANN model and Kriging model.The approximation accuracy of Support Vector Regression model is slightly highest among four surrogate models,which can more effectively replace multiphase flow numerical simulation model.Therefore,in the following optimization process,Support Vector Regression model is selected and used in optimization model for calculating the remediation efficiency of different remediation strategies.Surrogate model,which is built on the basis of adaptive sampling method,often referred to adaptive surrogate model.With the iterative sampling process of adaptive sampling method,the parameters of Support Vector Regression surrogate models are constantly retraining,the approximation accuracy of the surrogate model has been improved.(3)Considering the uncertainty of surrogate model,the probabilistic constrained programming optimization model of surfactant-enhanced DNAPLs-contaminated groundwater remediation is constructed with the remediation cost minimization as the objective function.And simulated annealing algorithm is used to solve the problem,compared with the results of deterministic results,and the reliability of the optimization design is also evaluated under uncertainty.The probabilistic constrained programming optimization model can give the optimal remediation designs under different confidence levels,which is more consistent with the objective reality system,and can give more information about the optimal remediation designs to decision makers.It has great theoretical significance and practical significance to explore new technology for the surrogate models of multiphase flow numerical simulation model,and also enrich the theories and methods of the optimization design of DNAPLs-contarminated groundwater remediation based on surrogate models.