Prediction of the removal of solid suspensions and chemical oxygen demand from a pharmaceutical wastewater plant using a neural network approach

Abstract

This study aimed to model the removal efficiency of chemical oxygen demand (COD) and solid suspension (SS) from a real pharmaceutical wastewater treatment plant (WWTP) using artificial neural network-multilayer perception (ANN-MLP). The ANN model was developed using experimental data which were collected during four years. The input variables of the neural network are water pH, temperature, SS, and COD. The percentages of removal of COD (CODRE) and of SS (SSRE) are considered as output variables. The Levenberg–Marquardt algorithm was utilized to train ANN. It was found that the ANN architecture has two hidden layers with 8 neurons. The results of ANN models were compared with the measured data based on the correlation coefficient (R2) and mean square error (MSE). It was noticed that the best ANN model provides good accurate results with R2 values of 0.9783 for SSRE and 0.9826 for CODRE, and a value of MSE equal to 1.695 10−3. This study may aid the users to adjust operational parameters in recovering COD and SS in the case of the process treatment of industrial effluents.