ANN Based Approach For Simultaneous Detection of Groundwater Pollution Origin and Release Characteristics

Abstract

Present study aims at simultaneously identifying the three essential features of groundwater pollution source identification (GPSI) problem viz. the pollutant source location i.e. the distance between origin and observation point (x), duration of the pollutant release (T₀) and its concentration strength i.e. the pollutant release flux history (C₀). Four Artificial Neural Network (ANN) models were developed, the three models, viz. ANN1, ANN2, ANN3 identified the three source parameters individually; while the fourth model ANN4 aimed at identifying the source parameters x, C₀ and T₀ simultaneously. The model input comprised multiple sets of concentration breakthrough curves (BTCs) at 50 discrete ‘x’ for 10 different T₀ and 10 varying C₀ values. Employing the governing partial differential equation of pollutant transport in groundwater, 383130 patterns of BTCs were generated for training and testing of ANN models. Model performance was evaluated using standard statistical measures to recognize the optimal ANN architecture. A network with architecture 14-15-3 was found to be optimum for ANN4 and resulted in a average absolute relative error (AARE) of 9.17, 19.96, 16.75 for identifying x, C₀ and T₀ respectively. Single output ANN models performed better than the multiple output ANN model. While comparing for individual source parameters as target variables, both the models (single and multiple outputs) could efficiently identify the source location. Whereas, the release concentration and duration of pollutant release estimation by single output model performed better than the multiple output model. The proposed ANN based solution of the GPSI problem illustrated here, has a practical application in formulating strategies for regulating and penalizing the agencies accountable for the groundwater pollution.