Optimization of photocatalytic oxidation reactor for air purifier design: Application of artificial neural network and genetic algorithm

Abstract

Photocatalytic oxidation (PCO) is a promising approach for eliminating volatile organic compounds (VOCs) from indoor environments. The generation of toxic by-products, however, prevents this method becomes widely used. There is a lack of comprehensive model for the prediction of PCO removal efficiency and by-products generation, and its optimization. To address that, this study focusses on modeling and optimization of PCO based on artificial neural networks and genetic algorithms. The concentrations of target VOCs and three common by-products are estimated using the ANN model, which takes seven input parameters into account. They include; PCO type, number of PCO layers, air velocity, relative humidity, light intensity, and target VOC and its concentration. A number of ANN-based estimation concentration models are trained, validated, and tested using the experimental data from bench and pilot scale set-ups (a total of 260 data points). The best performance is achieved with a three-layer feed-forward back-propagation structure with 11 neurons in the hidden layer (7:11:4), tansig-purelin transfer function, and Levenberg-Marquardt (LM) as the training algorithm. The proposed ANN model could be trained to predict the concentration of different VOCs and formed by-products with a high accuracy (R2 of 0.99), which was further validated and tested with 0.98 and 0.97 precision, respectively. The relative importance of input parameters in the removal of target VOC and generated by-products showed that PCO media and type of target VOC were the most influential variables. Further, multi-objective optimization for maximizing removal efficiency and minimizing by-products showed that the optimal condition for PCO process is achieved when two layers of glass fiber-based PCO media is used as a photocatalyst, and the system is operating at air velocity of 0.91 m/s and relative humidity of 48.7%, and butyraldehyde concentration of 47.3 ppb.