Adaptability Analysis Of Surrogate Models In Identifying Of Groundwater Contaminant Source Release History

With the rapid development of China's economy,water resources as the pillar of economic construction,its pollution problems are also increasingly serious,and the negative impact of groundwater contaminant is also increasingly exposed.Strengthening the treatment and restoration of groundwater contaminant has become one of the major problems related to people's livelihood.However,due to the complexity of the groundwater system and the difficulty in obtaining information on the source of contaminant,it is difficult to control groundwater contamina nt at the source.In addition,the cost of treating and repairing groundwater contaminant is high,time-consuming and difficult,and existing remediation measures cannot fully play its role.Therefore,in order to better manage groundwater contaminant and reduce the risk of groundwater contaminant,the inversion of the characteristics of groundwater contaminant sources has become the focus of research on groundwater contaminant.This paper mainly discusses the applicability of different types of surrogate models to the inversion of groundwater contaminant sources release history under complex hydrogeological conditions,and improves the inversion stability by constructing an ensemble surrogate model.Among the many inversion methods,the optimization algorit hm is the most common and efficient method to solve the groundwater inversion problem.Among them,genetic algorithm(GA)is the most common heuristic algorithm,which is applied to solve the problem of contaminant source inversion in two-dimensional aquifer,and found that the computational efficiency of the model can be improved.Using the two modules of MODFLOW and MT3 DMS in GMS software to simulate and predict the migration process of groundwater contaminant,the simulation model is established.The optimization model is combined with the simulation model to obtain the contaminant source release process through inversion.Although the use of genetic algorithms can improve the efficiency of calculation,it is still time-consuming to repeatedly call the simulation model in the inversion process,which makes the applicability of this method not strong.An alternative model can solve this problem.Although the use of surrogate models is an effective method to solve the inversion of groundwater contaminant sou rces,the basic principles of different alternative models are different,and there are also differences in the performance of the inversion process.This paper mainly studies the applicability of different surrogate models to the historical inversion of groundwater contaminant sources under complex hydrogeological conditions,and improves the inversion stability by forming a combination of alternative models.The main research contents are as follows:(1)Applicability of an individual surrogate models in the inversion of groundwater contaminant source release historyIn order to study the applicability of different types of individual surrogate models to the inversion of groundwater contaminant source release history under complex hydrogeological conditions,four typical surrogate models were selected in this paper: Kriging,Generalized Regression Neural Network(GRNN),Least Squares Support Vector Machine(LSSVM)and Nuclear Extreme Learning Machine(KELM).Three different cases of complexity and hydrogeological characteristics are set according to the boundary shape of the watershed,the groundwater flow state and the observed noise level.The results show that the Kriging surrogate model has the highest inversion accuracy among the three different aquifer s and different noise levels.The LSSVM surrogate model and the KELM surrogate model have good applicability in different aquifers.However,under the condition of non-stable flow heterogeneous aquifer,the applicability is poor for the observation of well observations with high noise level;while the GRNN surrogate model has higher accuracy in the aquifer of heterogeneous steady flow.But aquifers with heterogeneous unsteady flows are not suitable for use,and it is not possible to stably treat observational observations with observed errors.(2)Study on the weighting method of an individual surrogate models in an ensemble surrogate modelThe shortcomings of the view weighting method and the objective weighting method are the excessive dependence on the opinions of experts and the quantitative methods that rely too much on statistics or mathematics.Therefore,in order to obtain the weight value of the model s more reliably,the main method in this paper is to combine the subjective weighting method with the objective weighting method.In this paper,the objective weighting method is the entropy weight method,and the weight value is objectively obtained according to the output data of the surrogate models simulation stage.However,the objective weight value s show that there is no significant difference between the three individual surrogate models.Combined with the subjective objective weighting method selected by the foreground theory method in this paper,the comprehensive weight value is obtained.Using this combination weighting method to obtain weight values is more reliable.(3)Applicability of the ensemble surrogate model in the inversion of groundwater contaminant source release historyAlthough the individual surrogate models are more accurate when inverting the release history of groundwater pollution source,the inversion process is not stable.In view of the inefficiency of the simulation model,an ensemble surrogate model is proposed to improve the accuracy and robustness of the results.The ensemble surrogate model proposed in this paper consists of the following three learning models: GRNN surrogate model,LSSVM surrogate model and KELM surrogate model.Using the combination weighting method to form an ensemble surrogate model,the accuracy and robustness of the inversion results were tested not only on the stable flow homogeneous aquifer,but also on the unsteady flow heterogeneous aquifer.Similarly,under complex hydrogeological conditions,the inversion results of the observational values of the observational wells with different degrees of observation errors were tested.The results show that the proposed ensemble surrogate model is an effective method to solve the inverse contaminant source inversion problem,and the inversion results are n ot only accurate but also robust.