Bayesian Regularization-Trained Multi-layer Perceptron Neural Network Predictive Modelling of Phenol Degradation using ZnO/Fe2O3 photocatalyst

Abstract

Listen

Translate article

The processing of crude oil in the onshore platform often results in the generation of produce water containing harmful organic pollutants such as phenol. If the produce water is not properly treated to get rid of the organic pollutants, human exposure when discharged could be detrimental to health. Photocatalytic degradation of the organic pollutant has been a proven, non-expensive techniques of removing these harmful organic compounds from the produce water. However, the detail experimentation is often tedious and costly. One way to investigate the non-linear relationship between the parameters for effective performance of the photodegradation is by artificial neural network modelling. This study investigates the predictive modelling of photocatalytic phenol degradation from crude oil wastewater using Bayesian regularization-trained multilayer perceptron neural network (MLPNN). The ZnO/Fe2O3 photocatalyst used for the photodegradation was prepared using sol-gel method and employed for the phenol degradation study in a batch reactor under solar irradiation. Twenty-six datasets generated by Box-Behken experimental design was used for the training of the MLPNN with input variables as irradiation time, initial phenol concentration, photocatalyst dosage and the pH of the solution while the output layer consist of phenol degradation. Several MLPNN architecture was tested to obtain an optimized 4 5 1 configuration with the least mean standard error (MSE) of 1.27. The MLPNN with the 4 5 1 architecture resulted in robust prediction of phenol degradation from the wastewater with coefficient of determination (R) of 0.999.