Optimization Of Remediation Strategy Of DNAPLs-Contaminated Aquifer Based On Surrogate

Extensive use of petroleum products have resulted the petroleum organic pollutants penetrating into the aquifer,which have caused serious groundwater pollution,and threaten environment and drinking water safety.These liquid organic contaminants are nearly immiscible in water,and maintain separate phases in aquifer,which are generally known as non-aqueous phase liquids(NAPLs).NAPLs can be classified into two types:light NAPLs(LNAPLs)have densities lower than water;dense NAPLs(DNAPLs)have densities greater than water.DNAPLs-contaminated sites are difficult to remediate,and several conventional remediation techniques(e.g.,pump and treat and vapor extraction)are commonly unsuccessful or have limited effects in treating DNAPLs-contaminated aquifer.The main reason of the limited remediation efficiency is that DNAPLs have high density,strong toxicity,low solubility and high interfacial tension.Surfactant flushing technology,also known as surfactant enhanced aquifer remediation(SEAR),is an enhancement for pump and treat technology.SEAR has solubilization and mobilization to hydrophobic organics,which can enhance the mobility and solubility of DNAPLs in water phase.This significantly increases the remediation efficiency of the pump and treat technique.However,due to the high cost of SEAR process,it is an urgent as well as theoretically and practically important scientific problem to analyze the different influence factors have on the remediation efficiency,and to optimize and analyze the remediation strategy,which can improve the remediation efficiency and reduce the remediation cost.These goals can be achieved through comprehensively applying sensitivity analysis,multi-phase flow numerical simulation,and nonlinear programming based on field investigation.In this paper,aiming at nitrobenzene contaminated aquifer remediation problem,the remediation strategy of the contaminated aquifer was optimized and analyzed,through the applying of sensitivity analysis technique,multi-phase flow simulation model,surrogate model and mixed-integer nonlinear programming(MINLP)etc.Firstly,on the basis of multi-phase flow numerical simulation model,Sobol'global sensitivity analysis method was used to assess the contributions of the variables on remediation efficiency,and the variables that obviously influence the remediation efficiency was identified as the input variables.This sensitivity analysis process can reduce the number of variables and further reduce the computational burden of optimization process.Secondly,optimal Latin hypercube sampling method was adopted to collect input data in the feasible region for input variables of multi-phase flow simulation model,and output data were obtained through running of multi-phase flow simulation model.The samples coverage degree of the optimal Latin hypercube sampling method was compared with that of the Latin hypercube sampling method.Once more,Polynomial regression(PR),radial basis function artificial neural network(RBFANN)and Kriging methods were used at the same time to build surrogate model of multi-phase flow simulation model,and the accuracy of the three models were then compared,and the model with highest accuracy was selected for further use.At last,mixed integer nonlinear programming optimization model was constructed with the remediation cost minimization as the objective function.The developed surrogate model was embedded in the optimization model for replacing the input output relationship of the simulation model,which can reduce the computational burden of the model resolving process.Genetic algorithm(GA)and penalty function were used to solve the optimization model,and the optimal remediation strategy was obtained.This research will enrich the theoretical and technical connotation for the analysis of remediation strategy optimization of the aquifer contaminated by DNAPLs.General conclusions drawn from this study are the following:(1)Sobol' sensitivity analysis results demonstrate that:in these particular settings,remediation duration is the most important variable influencing remediation efficiency.followed by total extraction rate,while the surfactant concentration has smallest influence on remediation efficiency.Second-order and third-order sensitivity indices are all smaller than 0.05,which indicates that these three factors have limited interaction effects to remediation efficiency,and can be neglected to some extent.Therefore,the surfactant concentration can be fixed at a constant value(median value of its range),which can reduce the number of variables considered in the optimization process,and further reduce the computational cost of the optimization process.(2)After optimization,the center L2 deviation decreases significantly.Compared with Latin hypercube sampling method,the introduction of optimal Latin hypercube sampling method improves the coverage degree of samples significantly.(3)Both RBFANN model and Kriging model have acceptable approximation accuracy.The approximation accuracy of the Kriging model is slightly higher than that of the RBFANN model.PR surrogate model has the lowest approximation accuracy in these three surrogate model.This is because polynomial function is limited in representing the nonlinear problem,especially the high-order nonlinear problem.Therefore,in the following optimization process.Kriging model is selected and used in optimization model for assessing the remediation efficiency of different remediation strategies.(4)The MINLP model is constructed with the remediation cost minimization as the objective function.The developed surrogate model is embedded in the optimization model for replacing the input output relationship of the simulation model,which can reduce the computational burden of the model resolving process.Genetic algorithm(GA)and penalty function are used to solve the optimization model,and the optimal remediation strategy is obtained.The introduction of mixed integer nonlinear programming effectively resolves the optimization problem of both wells location and wells rates.(5)In the simulation optimization problem solving process,using surrogate model to replace the simulation can considerably reduce the computational burden,and keep a good accuracy.