Development of Unknown Pollution Source Identification Models Using GMS ANN–Based Simulation Optimization Methodology

The detection of groundwater pollution sources is an important but a very difficult task. The location and magnitude of groundwater pollution sources can be identified using inverse optimization technique. The technique is also known as simulation-optimization approach where the aquifer simulation model is incorporated with the optimization model for finding the unknown pollution sources in an aquifer. The efficiency of the simulation-optimization model is highly related to the performance of the simulation model. This study develops three improved methodologies for identification of unknown groundwater pollution sources. In the first approach, the groundwater modeling system (GMS) is linked with the optimization model for solving source identification problem. The optimization model is solved using the direct-search method. The incorporation of GMS with the optimization model allows for the solving of a bigger real-world pollution source identification problem. The challenge of this approach is the linking of the external simulator GMS with the optimization model. This has been overcome by executing GMS in the Matlab environment. The main drawback of the approach is that the approach is computationally extensive. For reducing the computational time, the second approach uses the artificial neural networks (ANN) model to simulate the flow and transport processes of aquifer. The ANN model is then externally linked with the optimization model. This approach drastically reduces the computational time of the simulation-optimization model. The problem that was solved in few days can now be solved in a few hours. However, most of the time, it yields only the near optimal solution. Therefore, in the third approach, a hybrid optimization approach is presented that initially solves the problem using ANN-based simulation-optimization model. The solution obtained by the ANN-based model is then used as the initial solution for the GMS-based model. This approach is computationally more efficient than the GMS-based approach and also more accurate than the ANN-based model. The efficiency and accuracy of the proposed approaches are demonstrated using two illustrative study areas.