Identification Of Groundwater Pollution Sources Based On Kriging Alternative Model And Improved Bayesian-MCMC Methods

With the increase of human activities, groundwater pollution became more and more severe and has been a serious threat to the drinking water safety and the ecological environment. Therefore, it is of great significance to make a scientific and reasonable groundwater restoration and treatment plan to effectively cope with the groundwater pollution.The necessary basis of making a reasonable restoration plan is to find out the characteristics of the pollution sources and the spatial-temporal distribution of the pollutants. Meanwhile, these findings can also provide a basis for the risk evaluation on the groundwater pollution, and provide a criterion for the responsibility for the pollution. The inversion identification research on the groundwater pollution sources has great significance in restoring and improving the groundwater environment, realizing a harmonious relationship between human and nature and promoting the green development of the economy and the society.Currently, the widely used deterministic inversion method such as optimization algorithm can only provide an “optimal solution” or a“point estimation” for the inverse problem rather than take adequate consideration on the indeterminacy of the mathematical model, model parameters and observed data. However, the maximum solving difficulty of groundwater pollution inversion problem is the ill-posed problem caused by the indeterminacy of the observed data. Hence, the optimization algorithm has its limitation in inversion problem and is unable to adapt the characteristics of the inversion problem.This paper adopted the statistical inversion method(or known as probability inversion method) of Bayesian inference to conduct the inversion identification research on groundwater pollution sources. Based on probability theory, Bayesian inference processed variables into random variables, with all the indeterminacy expressed in the form of probability distribution, and took full advantage of prior information. The method can describe and express the indeterminacy in the inversion identification problems of water pollution sources, meet the inversion characteristics of groundwater pollution sources, and provide solutions for the inversion problem in the form of probability distribution. It is more informative, objective, reliable and accurate.Based on conventional MCMC-Bayesian algorithm, this paper added step transformation, which was means the step changed with the distance between the sample value and the true value: when the distance is far from the truth, the large step size was used to search, while the small step size was used to search in the vicinity of the true value. This improvement was to further increase the accuracy, efficiency and stability of the method.On the basis of two-dimensional groundwater steady-state flow and two-point source pollution discharge in separated periods, this paper comprehensively applied the analytic methods including groundwater solute transport numerical simulation model, substitution model and improved MCMC-Bayesian, and studied the groundwater pollution inversion identification based on Kriging substitution model and improved MCMC-Bayesian method.First, this paper established a groundwater solute(pollutant) transport numerical simulation model, used Latin hypercube sampling method and optimal Latin hypercube sampling method for the sampling in feasible regions, and analyzed the covering sizes of the two above methods on the sampling space. Then, the paper substituted the sampling results into the groundwater solute transport numerical simulation model and obtained related input value after operated the model. Based on the input-output sample data set, the paper studied and applied RBFNN(Radial Basis Function Neural Network) and Kriging method respectively to establish the substitution model of the groundwater solute transport numerical simulation model, comparatively analyzed the approximation accuracy of the two substitution models on the simulation model, and selected the optimal modeling method of the substitution model. On the basis of the higher precise substitution model, this paper employed the classical MCMC-Bayesian method with the improved MCMC-Bayesian method to conduct inversion on the pollution discharge level of the two pollution sources at each period of time, and discussed and evaluated the comparatively analyzed results.Based on the studies, this paper mainly concluded:(1)Compared with the Latin hypercube sampling method, the optimal Latin hypercube sampling method can effectively improve the cover degree of the sample points on the sampling space.(2)In the inversion identification calculation process of groundwater pollution sources, the use of substitution model as the conversion of simulation model to replace the simulation model greatly reduced the calculation load and maintained a good accuracy. And the approximate accuracy of Kriging method is even higher than that of RBFNN.(3)Using MCMC-Bayesian to study the inversion identification of groundwater pollution sources, the accuracy of the inversion results is high and stable.(4)The addition of step transformation in MCMC-Bayesian method can substantially improve the calculation accuracy, efficiency and stability of the MCMC-Bayesian method.