Groundwater Model Approximation with Artificial Neural Network for Selecting Optimum Pumping Strategy for Plume Removal

Cleanup of contaminated aquifers by pumping and injection is one of the commonly used approaches for the remediation of groundwater contamination. Contaminant transport travel time in groundwater can be calculated using a method called ‘particle-tracking’ based on advection. The travel time of the contaminants is a highly non-linear and nonconvex function of pumping/injection rates and well locations. Global optimization (GO) techniques are therefore appropriate for finding an optimum pumping strategy. However, a pronounced disadvantage of these techniques is that they require running simulation models – in this case groundwater flow and particle tracking models quite many times taking very long time to find an optimal solution. On the other hand, Artificial Neural Networks (ANN) are nowadays one of the widely used modelling techniques which can approximate a non-linear relationship between input and output data sets without considering physical processes and the corresponding equations of the system. As a result, an ANN model is much faster than a physically based model which it approximates. In this study, ANNs were trained to approximate the groundwater models MODFLOW and MODPATH using the data generated by these models. The resulting ANN models were then coupled with a GO tool, GLOBE, to find optimal pumping strategies. The experiments were carried out using different number of pumping wells and different GO algorithms.