Simulation of Contaminant Transport through Saturated Media using Visual Modflow and ANN

The contaminant transport processes through saturated media are highly complex and non-linear. These processes must be simulated to evolve an optimal strategy for control and remediation of ground water pollution in an aquifer. Incorporation of a very large and complex simulation model within classical optimization framework becomes very difficult, and computationally infeasible The main aim of this study is to evaluate the use of a trained Feed forward Back propagation and Radial Basis Function networks to simulate contaminant process through saturated media. The training and testing data sets were generated using the finite difference scheme.

finite difference scheme. Analysis was done for one different hypothetical study area and one real study area. In the first scenario, there was single spatial location of source and 3 observation wells. In the second scenario, there was single spatial location of source and 3 observation wells. In both the scenarios training and validation using ANN was done separately on each well. For the first scenario, Breakthrough Curve Data sets are generated using the Cranck-Niconlson finite difference scheme. For the second scenario, Breakthrough Curve Data sets are generated using the Visual Modflow.

The training and testing results show that for both hypothetical and real study areas, the trained RBF network is capable of simulating transport process satisfactorily and performing superior than FFBP network. The simulation error in terms of the mean square error, average absolute relative error and correlation coefficients between the RBF simulation results and numerical simulation results are satisfactory. The evaluation of the proposed methodology suggests that a properly trained RBF network is potentially useful for computationally efficient simulation of the complex pollution transport process through saturated media.