Application of artificial intelligence for predicting reaction results in advanced oxidation processes

Abstract

Response surface methodology (RSM) was used to predict total organic carbon (TOC) removal in advanced oxidation processes in wastewater-treatment fields. Because the disadvantage of RSM was poor generalization ability, we experimented with applying an Artificial neural network (ANN) to the processes. From RSM experimental data results, we found that ANN had better prediction ability than RSM. In catalytic wet air oxidation, the TOC-removal Mean square error (MSE) based on an ANN testing set was 6.10, which was smaller than that based on RSM (11.89), and the Isophorone-conversion MSE based on an ANN testing set was 0.31, which was also smaller than that based on RSM (3.13). In catalytic wet peroxide oxidation, the RSM-predicted TOC removal was 81.67%, which was not as close to the experimental TOC removal (91.4%) as the 88.40% attained by ANN. In wet electrocatalytic oxidation, the ANN-based prediction of TOC removal was 90.46%, which is closer to the experimental value (88.88%) than RSM-based 92.13%, whereas ANN-based prediction of chemical oxygen demand removal was 93.81%, which was closer to the experimental value (91.82%) than the RSM-based 94.33%. In conclusion, ANN had great potential for the future of chemistry and the chemical industry.