Groundwater Pollution Source Identification Based On Random Statistical Methods

The increasing intensity of human activities has led to more serious groundwater pollution,which has posed a threat to the safety of drinking water and ecological environment.Groundwater pollution is characterized with hidden existence,hysteretic nature,which makes it difficult to design groundwater restoration schemes,evaluate pollution risks and determine responsibilities for pollution.Groundwater pollution source identification is a prerequisite for formulation of practical restoration schemes and is a necessary condition for evaluation of groundwater pollution risks and for determination of pollution responsibilities.The inverse identification of groundwater pollution sources refers to the use of finite and discrete groundwater observation data to solve the mathematical simulation model by means of inversion algorithm,in order to identify the numbers,locations and release history of pollution sources.Inverse identification of groundwater pollution sources,as a typical inverse problem,is very complex due to its ill-posed nature,that is,the solution is non-unique,non-existent and unstable.At present,relevant research on inverse identification of groundwater pollution sources is still in the stage of development There are various factors influencing the effect of inverse identification of groundwater pollution sources,including selection of inverse identification methods,other factors such as errors in water quality observation data,abnormal measurement values,and locations and numbers of new monitoring wells.Therefore,on the basis of field investigation,how to analyze the impacts of various factors on the inverse identification results,to analyze and optimize the layout schemes of new monitoring wells in order to improve the accuracy and efficacy of inverse identification of pollution sources by joint use of random statistical methods,simulation modelling of solute transport,and integral programming models,has become a concern with great theoretical and practical significance that calls for prompt solution.In this dissertation,four hypothetical situations were set up to study the inverse identification problem of groundwater pollution sources.Based on a case study of pollution source identification,this paper has conducted research into the problem of inverse identification of pollution sources by means of four methods,namely solute transport simulation model,surrogate model,random statistical method(with adjoint state method and Bayesian method),multivariable control method and 0-1 integer programming model,in order to identify features of groundwater pollution sources(number,location and release history).Firstly,according to the hydrogeological condition of hypothetical examples,a numerical simulation model of solute transport was established by using the historical water levels and water quality observation data,the number and location of pollution sources were initially identified through solution of inverse problem by adjoint state method.Within the possible range of pollution source intensity,stratified sampling was conducted by Latin hypercube sampling method.After running the simulation model,corresponding output response was obtained.Based on input-output sample data set,a surrogate model was set up by Kriging method to replace the simulation model,and parameter optimization was performed by using adaptive weighting particle swarm optimization algorithm to obtain the best surrogate model.It is necessary to call repeatedly simulation model to identify pollution sources in Bayesian method,however,replacement of simulation model with a surrogate model can greatly reduce computational load and maintain fairly good accuracy.When implementing the surrogate model with the Bayesian method in identifying the pollution source intensity,MCMC is used to solve the posterior probability to obtain a statistical estimation of the pollution source intensity.On the basis of the previous research,a multivariable control analysis of the effect of relevant factors on pollution source identification was performed,such as the number and location of monitoring wells,error and abnormal values of water quality observation data.Finally,a 0-1 integer programming optimization model was constructed by maximizing the coverage by new monitoring wells in an area with a greater degree of pollutant concentration as the objective function.Solution to the optimization model was obtained by implicit enumeration method to obtain optimal layout scheme for new monitoring wells.With water quality observation data from new and previous monitoring wells as feedback information,and re-identification of pollution sources by random statistical method constituted a feedback correction iteration.Such iteration was repeated until convergence between the measured value of pollutant concentration in monitoring wells and the simulated value was achieved.The converged value was the solution to the inverse problem.The major conclusions can be summarized as follows from the preseat study:(1)The number,location and release history of pollution sources can be identified through solving the inverse problem by combining adjoint state method and Bayesian method.The computations in four different situations indicate that the above-mentioned two methods fit in well with features of different aquifers(homogeneous and heterogeneous)and pollution sources with different ways of emission(continuous emission and divided-period emission).(2)An assessment of the approximation of a surrogate model to a simulation model was performed by using the following five indices,namely deterministic coefficient R2,mean absolute error,mean relative error,maximum absolute error and maximum relative error.Results showed that Kriging model is a close approximation to simulation model.During inverse identification of pollution sources based on Bayesian method,a surrogate model was an alternative to a simulation model;replacement of a simulation model with a surrogate model greatly reduced computational load and also maintained a fairly high degree of accuracy.(3)Analysis of contributory factors in identification of groundwater pollution sources show that adjoint state method still works well in identifying the number and location of pollution sources in the following two situations,that is water quality observation data at a low level of error(б=0.01)and abnormal numerical value of water quality observation data in a single monitoring well at a given time.Data missing has such a significant impact on identifying through inversion the number and location of pollution sources by adjoint state method that it is possible to obtain an erroneous solution to an inverse problem;the number and location of water quality monitoring wells has a great impact on the identification accuracy by adjoint state method and it is possible to improve the accuracy of inverse identification by increasing the number of wells or optimizing the location of monitoring wells in a reasonable way.(4)Water quality observation data at a low-error level(б=0.01)has a certain influence on the accuracy of inversion solutions by Bayesian method.The inversion error increases with the increase of error levels.The existence of abnormal water quality observation data in a single monitoring well at a given time is likely to cause Markov chain not to converge,that is,inability to obtain an inversion solution,for Bayesian method,which takes likelihood function as a normal distribution;Bayesian method with missing data has a low-level of inverse identification,and even a negative value may arise;the number and location of water quality monitoring wells has a great impact on the identification accuracy by Bayesian method and it is possible to improve the accuracy of inverse identification by increasing the number of wells or optimizing the location of monitoring wells.(5)An integer programming optimization model with 0-1 variables was constructed by maximizing the coverage with new monitoring wells in an area with a greater degree of pollutant concentration as the objective function,and using controllable factors during designing new monitoring wells as decision variables;then,the optimization model was resolved by implicit enumeration method to obtain the best layout scheme for new monitoring wells.The accuracy of normalized error(NE)obtained by increasing the number of monitoring wells decreased by 0.15%compared with that without doing so.Therefore,this dissertation has proposed a feedback correction iteration strategy for pollution source identification,based on optimized layout of monitoring wells and random statistical method;it was completed by optimizing the layout of new monitoring wells,feeding water quality observation data back to the pollution source identification process so as to modify source identification results,and to iterate until a satisfactory solution was obtained.Application of 0-1 integer programming solved the problem of the optimal layout of new monitoring wells,which can be combined with source identification method with the aim of improving the accuracy of pollution source identificatioin.This dissertation has explored a new,practical and effective technique for the identification of groundwater pollution sources,which possess important theoretical and practical significance.