Modeling the effect of process parameters on the photocatalytic degradation of organic pollutants using artificial neural networks

The need for pollutant-free wastewater has necessitated a huge volume of research on the photocatalytic degradation of organic pollutants. The data obtained from various photocatalytic degradation experimental runs can be employed in data-driven machine learning modelling techniques such as artificial neural networks. In this study, the use of Levenberg-Marquardt-trained artificial neural network for modelling the photocatalytic degradation of chloramphenicol, phenol, azo dye, gaseous styrene, and methylene blue is presented. For each of the photocatalytic degradation processes, 20 neural network architectures were investigated by optimizing their hidden neurons. Optimized ANN configurations of 3−20-1, 3−5-1, 3−2-1, 4−17-1, 4−6-1, and 3−10-1 were obtained for modelling the photodegradation of chloramphenicol, phenol, phenol, azo dye, gaseous styrene, and methylene blue, respectively. The optimized ANN architectures were robust in predicting the degradation of the organic pollutants with R2 > 0.9 at a 95 % confidence level with very low mean absolute errors. The sensitivity analysis using the modified Garson algorithm revealed that all the process parameters significantly influenced the photodegradation of the organic pollutants. The photocatalyst concentration, phenol concentration, pH of the solution, hydrothermal temperature, and methylene blue initial concentration were however found to have the most significant influence on the photodegradation processes. The ANN algorithm can be implemented in a photocatalytic degradation process for making vital decisions regarding the operation of the process.