Modeling the effect of operating variables on sorption of PAHs from aqueous streams onto orange peels using artificial neural network pattern recognition.

Abstract

The application of artificial neural network (ANN) technology to the simulation of factors affecting sorption of polycyclic aromatic hydrocarbon (PAHs) onto ripe and unripe orange peels is presented in this work. A 3-layer backward propagation network structure was applied using pattern recognition tool in MATHLAB 7.9.0 (R2009). Optimum number of neurons in the hidden layer used was 20 with MSE value of 0.000912. Parameters such as contact time, adsorbent dosage, pH, and particle size were used as input variables while the output of the ANN was the pollutant removal concentration. The study showed that neural network pattern recognition generated data for pollutant removed using ripe and unripe orange peels agreed to a large extent with the laboratory data. The regression correlations obtained for both ripe and unripe orange peels closely approximated to 0.99. In general, the result of this study indicated that particle size was the most significant factor that affects sorption of PAHs. The other factors considered in this study affected the sorption of PAHs in the order, contact time> pH > adsorbent dosage.

Keywords: ANN, Backward propagation, MSE, Effluent concentration, Pollutant removal efficiency.