Inversion Identification Of LNAPLs Contamination Source In Groundwater Based On Stacking Surrogate Model

Oil products often leak into the groundwater environment due to improper disposal or accidents in the production,storage and transportation,etc.,which will cause pollution to groundwater.Because groundwater is buried and moved below the surface of the ground,groundwater pollution has the hidden and found lag,which leads to the lack of understanding and mastery of groundwater pollution sources.This is not conducive to the responsibility identification of pollution,risk assessment and the design of repair plan and other follow-up work.Therefore,it is necessary to study the analysis of the oil pollution sources in groundwater.The analysis and research of groundwater pollution source is a process of determining the relevant information(including the location and release history of the pollution source)of groundwater by using the dynamic observation data of groundwater(including groundwater level and pollutant concentration),combined with the auxiliary information such as relevant investigation data of the research area.The problem of groundwater pollution tracing is a typical inversion problem in mathematics,which has the characteristics of nonlinearity and discomfort.Unlike the forward problem,the information to be solved in inversion is often more than known information,which makes it more difficult to solve the inversion problem.In this paper,BP-Stacking surrogate model,cycle update process(CUP)strategy and simulation optimization method are used to study the pollution source analysis of light non-aqueous phase fluids(LNAPLs)by combining hypothetical examples with practical applications.Firstly,the geological and hydrogeological conditions of the research area are generalized and the conceptual model is established.Then,the reasonable range of the data and parameters of the simulation model is given,and the initial estimation is given.The multi-phase flow model of LNAPLs pollution in groundwater is established.Secondly,the variables to be identified are determined and the surrogate model of multiphase flow numerical simulation model is established.The parameters of the simulation model are selected by local sensitivity analysis method,and the parameters of the simulation model with high sensitivity are taken as the variables to be identified together with the information related to the pollution sources.The Latin hypercube sampling method is used to sample the value range of variables to be identified,and the input data of the simulation model are obtained.Then,the simulation model is used to calculate the corresponding output data,namely,the training sample and the test sample.The training and evaluation accuracy of the stacking substitution model based on BP neural network are carried out.The applicability of BP-Stacking surrogate model is analyzed by comparing with krig and kelm.Finally,the information of pollution sources and the parameters of simulation model are identified.The loop update process(CUP)policy is constructed.In the CUP strategy,two simulation optimization processes are designed,which identify the sensitive parameters and the information of pollution sources respectively,and connect the two simulation optimization processes into a cycle loop.The hybrid homotopy difference evolutionary whale algorithm(HA-DE-WOA)is used to solve the optimization model.In order to reduce the calculation time and load,BP-Stacking surrogate model is embedded into the optimization model as an integral part of the optimization model.By identifying the pollution source characteristics and simulation model parameters,the identification results of the two simulation optimization processes can be promoted and updated continuously until the optimal solution of the problem is solved,namely the identification value of the variables to be identified.Through the above research,the following conclusions are obtained:(1)In the fitting accuracy of input-output in the simulation model,the accuracy of BP-Stacking surrogate model is higher than that of KRG and kelm.The BP-Stacking model based on stacking integrated learning and BP neural network is more suitable for solving the practical problems of multi-variable,high-dimensional and complex nonlinear mapping.(2)This study explores the effective solution of nonlinear optimization model.The homotopy algorithm(HA)and differential evolution algorithm(DE)are cited to the traditional whale optimization algorithm(WOA),which can solve the problem that the traditional WOA algorithm relies on the initial value selection and is prone to sink into local optimal.The hybrid homotopy differential evolution whale algorithm(HA de WOA)is constructed.Ha de WOA algorithm can search global optimal solution more efficiently and improve the accuracy of analysis results of LNAPLs pollution source in groundwater.(3)This study explores an effective strategy for solving the problem of groundwater pollution source analysis.The CUP strategy consists of two simulationoptimization processes,one is the identification process of pollution source related information,the other is the process of identifying the parameters of simulation model.In the CUP strategy,the identification results of the previous identification process are replaced into the latter identification process to update the identification results of the variables to be identified.Through several iterations,the identification results of the identification of LNAPLs pollution in groundwater are obtained.Compared with the results of simultaneous identification of pollution source information and simulation model parameters,the results obtained by using CUP strategy are more close to the true value of variables to be identified.Therefore,the CUP strategy can improve the accuracy of the identification results of LNAPLs trace in groundwater.