Artificial neural networks: development and application in groundwater pollution remediation design

Artificial neural networks (ANNs) are investigated as a tool for the simulation of contaminant loss and recovery in three-dimensional (3-D) heterogeneous groundwater flow and contaminant transport modelling. These methods have useful applications in expert system development, knowledge base development and optimization of groundwater pollution remediation. Conventional numerical model runs are used to develop the ANNs. ANNs have been analysed with the goal of estimating objectives that normally require the use of traditional flow and transport codes such as recovered mass, unrecovered mass and remediation failure. The inputs to the ANNs are variable pumping withdrawal rates at fairly unconstrained 3-D locations. A forward-feed backwards error propagation ANN architecture is used. The significance of the size of the optimization data set, network architecture and network weight optimization algorithm, with respect to the estimation accuracy and objective are shown to be important. Finally, cross-validation techniques quantify the quality of the weight optimization for strongly under described systems.