Application of a neural network model for the containment of groundwater contamination

Industrialisation is causing the environment and in particular groundwater to be affected by pollution. It is therefore imperative to adopt remediation techniques to control this contamination. In this study, the hydraulic containment method using extraction wells was adopted as the remediation technique. An optimisation model is developed to minimise costs of pumping used for the containment of groundwater contamination. The output from this is used to train a neural network model that has been developed for optimal evolution of pumping strategies. Neural networks are proving to be useful decision-making tools because they are able to store knowledge and can consider nonlinear relationships, fuzzy relations, etc. The optimisation model developed and the neural network model is applied in a case study. The feed forward neural network is adopted with the input nodes storing the water levels at the wells (five observation wells and one pumping well are considered) and the output node storing the optimal pumping rate for these water levels, which is obtained using the optimisation model. This neural network is trained with six input nodes, one output node and eleven nodes in the hidden layer. This neural network is trained with 45 patterns and tested with four patterns. The trained neural network proved to be very useful in making decisions on the number of pumping wells, and in obtaining the optimal pumping rate for each well. The user, on specifying a set of inputs (the water levels in the wells) to the network, can obtain the optimal pumping rate at all the extraction wells in order to ensure that the contaminant plumes have been contained within the specified area. © 2005 EPP Publications Ltd.