A neural network approach to predict the performance of recycled concrete used in permeable reactive barriers for the treatment of acidic groundwater
This study developed a neural network model for examining the performance of recycled concrete for the treatment of acidic groundwater. Concentrations of Al, Fe and Ca and alkalinity of the effluent were selected as the output parameters to simulate the performance of recycled concrete for neutralizing acidic groundwater. The input variables were the number of pore volumes, pH, oxidation reduction potential and the average hydraulic conductivity. Of the 658 experimental datasets available, 409 datasets were used for training, 184 datasets were used for validation, and the remaining datasets were used for cross-validation. The reported results indicate that the neural model is a valuable tool to assess and simulate the performance of recycled concrete. The sensitivity study confirmed that the selected input signals of the output estimate were equally important. A similar model could also be used for full-scale permeable reactive barrier installation provided that up-scaling issues such as the possible non-homogeneous nature of the recycled concrete and variation in groundwater quality can be effectively resolved.