Application of neural network approach for modelling COD reduction from real refinery effluent by electrocoagulation

Abstract The present study aims to investigate the feasibility of implementing the electrocoagulation (EC) process to treat Algiers refinery effluent. The electrocoagulation was performed by using scrap aluminum plate electrodes in monopolar-parallel mode. Several parameters, namely current density, reaction time, the electrolyte dose, and the initial chemical oxygen demand (COD) concentration were studied. The maximum removal of COD achieved was found to be 78.55%. Operating conditions at which maximum COD removal efficiencies were achieved at current density 8 mA/cm2, electrolyte dose 1 g/L, with 360 mg/L of initial COD concentration at working time of 40 min. An artificial neural network (ANN) was also utilized to determine predicted responses using neural networks for the 4-10-1 arrangement. The responses predicted by ANN were in alignment with the experimental results. The values of the determination coefficient (R2 = 0.978) and the root mean square error (RMSE = 21.28) showed good prediction results between the model and experimental data. Hence, the ANN model as a predictive tool has a great capacity to estimate the effect of operational parameters on the electrocoagulation process.