Use of artificial neural networks for modeling tds in groundwater in the berrahal region-annaba (ne Alegria)

In this research, we created a model for predicting total dissolved solids variables in groundwater wells as a function of groundwater quality status variables in the Berrahal region, such as temperature (T), pH, electrical conductivity (EC), sodium (Na⁺), potassium (K⁺), bicarbonates (HCO₃⁻), sulphates (SO₄^2-), and copper (Cu²⁺). Artificial neural networks were used to approximate the relationship between these different variables. The performance of two artificial networks were evaluated to determine which would be most effective in predicting total dissolved solids (TDS) concentrations in groundwater wells in the Berrahal region, the multilayer perceptron (MLP) and radial basis function (RBF) neural networks. Prediction results show that the neural network approach has good and wide applicability for modelling TDS in groundwater wells in the Berrahal region. The comparison between the model values by ANNs with experimental data reveals that MLP (BFGS 13) model with thirteen neurons in hidden layer provides accurate results (R²= 0,996, RMSE = 26.831 mg/l, r = 0.998 for the testing; R²= 0,992, RMSE = 46.706 mg/l, r = 0.996 for the training and R²= 0,961, RMSE = 64.861 mg/l, r = 0.980 for the validation).