| In the production,storage and transportation process,petroleum products may leakage into groundwater system,and causes groundwater pollution.groundwater pollution has the characteristics of hiding existence and finding hysteresis.Accurate characteristics information of groundwater pollution source is the premise of responsibility confirmation,risk assessment,and remediation of groundwater pollution.Therefore,it is pretty essential for above works to conduct the characteristic identification of groundwater pollution source.The accurate identification of pollution source characteristics depends largely on the amount of effective information(known conditions)provided by monitoring data.With the limitation of cost budget,the quantity of monitoring wells and the monitoring frequency are limited;moreover,there is a conflicting relationship between the former and the latter.Therefore,rational design of well quantity,location and monitoring frequency is significant to obtain more known conditions of pollution source identification and reduce monitoring cost at the same time.Under the simulation-optimization framework,The optimal design of groundwater monitoring scheme is based on the prediction results of simulation model,in which the groundwater pollution source is a source item of the simulation model;the accuracy of source characteristic information is crucial to the optimal design of groundwater monitoring scheme.Therefore,the identification of groundwater pollution source characteristics and optimal design of groundwater monitoring scheme are mutually feedback processes,it is necessary for accurate identification of groundwater pollution sources to conduct integrated study of two processes.The integrated study of pollution source identification and monitoring scheme optimal design focused on the single-phase flow problem in hypothetical examples at present,while there are few studies that focused on the multi-phase flow problem of actual sites.Moreover,the current studies on the monitoring scheme optimal design based on the simulation-optimization method are mostly focused on the optimization of quantity and location of monitoring wells,which can merely provide the static monitoring sampling scheme at one moment,and not simultaneously consider the optimization of monitoring frequency.therefore,a chemical plant where is located in Jilin City,Jilin Province is as study area.This study focuses on benzene pollution of study area(benzene is one kind of light non-aqueous phase liquids),this study comprehensively applies groundwater multi-flow numerical simulation,artificial intelligence algorithm,information entropy,multi-objective optimization and other theories and methods.The integrated study of pollution source identification and monitoring scheme optimal design was conducted,the main conclusions are follows:(1)In the process of source characteristic identification and monitoring scheme optimal design,Kriging,GA-SVR,and GA-BPANN methods were used to construct surrogate model of groundwater multi-phase numerical simulation model.In the stage of preliminary source characteristic identification and monitoring scheme optimal design,the coefficient of determination(R~2)of GA-SVR surrogate model were 0.981and 0.996 respectively for test samples.And the mean relative error(MRE)of GA-SVR surrogate model were 7.72%and 4.38%respectively.the fitting accuracy of GA-SVR surrogate model is far higher than that of other surrogate models,it can reliably and efficiently reflect the input-output relationship of simulation model,but the trainning time is longer.Therefore,GA-SVR surrogate model was used to couple with optimization model in two study stage.(2)On the basis of spare monitoring data,the groundwater LNAPL pollution source in study area was identified by simulation-optimization method.The fitting results between observed concentration and identified concentration showed that the mean absolute error of three existing monitoring at 6 monitoring moments was 131.2μg/L,the mean relatively error was 22.27%,the fitting accuracy was low indicated the known conditions provided by the existing monitoring data were insufficient,and the preliminary identification results of source characteristics was still uncertain.(3)According to the preliminary identification results of pollution sources,the number and location of added wells were designed.A 0-1 integer multi-objective programming model was constructed with the objective function of maximizing spatial-temporal information of monitoring data and minimizing monitoring cost.The multi-objective model was solved by NSGA-II,23 Pareto optimal monitoring schemes,existing monitoring scheme,and 23 random monitoring schemes identified5 LNAPL pollution source situations together.The results showed that:(1)for the overall identification accuracy,the optimal monitoring schemes were superior to the existing monitoring scheme and random monitoring schemes from the identification accuracy of source characteristics and the fitting accuracy of benzene concentration in monitoring wells.(2)for the local identification accuracy,some optimal monitoring schemes obtain higher identification accuracy with fewer monitoring wells and monitoring frequency than that of random monitoring schemes.(3)for monitoring cost,with the increase of the monitoring cost of the optimal groundwater monitoring scheme,the identification accuracy of source characteristics kept increasing,but the increasing is slower and slower.The identification accuracy of optimal monitoring schemes was higher than that of random monitoring scheme when the monitoring cost is equal.The three results indicated that the identification accuracy of source characteristics can be improved and the monitoring cost can be reduced by optimal design of well quantity,locations and monitoring frequency.In actual monitoring work,the decision of monitoring scheme shall be comprehensively considered according to the site overview,public health and safety,cost budget and etc. |
