Prediction model development for petroleum refinery wastewater treatment

Multi-stage biological treatment of petroleum refinery wastewater using different biological conditions (anaerobic–anoxic–aerobic) has many advantages over other biological methods. It can result in maximum treatment for type of complex wastewater. In this study, raw data obtained from two multi-stage biological reactors (MSBR) used for treatment of different loads of petroleum refinery wastewater was used for developing mathematical model that could predict the process trend. The data consists of 160 entries and were gathered over approximately 180 days from two MSBR reactors that were continuously operated in parallel. A Matlab code was written with two configurations of artificial neural network. The configurations were compared and different number of neurons at the hidden layer were tested for optimum model that represent the process behavior under different loads. The tangent sigmoid transfer function (Tansig) at hidden layer and a linear transfer function (Purelin) at output layer with 6 neurons were selected as the optimum best model. The model was then used for prediction; highest removal efficiency observed was 98% which was repeatedly recorded for various loads. Effluent concentration below 100 mg/L as chemical oxygen demand (COD) was recorded for influent concentration ranged between 900 and 3600 mg COD/L.