Backpropagation neural networks modelling of photocatalytic degradation of organic pollutants using TiO2-based photocatalysts

BACKGROUND The advanced oxidation process using photocatalysts has been proven to be an efficient technique used for the degradation of organic pollutants in wastewater. However, there exists a nonlinear relationship between the process parameters of the photodegradation reaction, which needs to be well understood for the design of an efficient photoreactor. This study employed a backpropagation artificial neural network (BPANN) for the modelling of photocatalytic degradation of indole, anthraquinone dye and methyl blue using undoped and Ag+-doped TiO2 catalysts. RESULTS A Levenberg-Marquardt algorithm was employed to train the BPANN by varying the hidden neurons to obtained an optimized architecture. Optimized architectures with 3-14-1, 4-12-1 and 3-16-1 consist of the input layers, hidden layer and the output layer, were obtained using the datasets from photodegradation of indole, anthraquinone dye and methyl blue, respectively. The optimized BPANN accurately predicts the indole, anthraquinone dye and methyl blue degradation as a function of colour removal from the wastewater. High coefficients of determination (R-2) of 0.999, 0.961 and 0.993 were obtained for the prediction of the photodegradation of indole, anthraquinone dye and methyl blue, respectively, with over 95% confidence level. The study revealed that dye concentration, catalyst dosage and reaction time have the highest level of importance for the photodegradation of indole, anthraquinone dye and methyl blue, respectively. CONCLUSION This study has demonstrated the robustness of BPANN for predictive modelling of photodegradation of organic pollutants such as indole, anthraquinone dye and methyl blue. (c) 2020 Society of Chemical Industry