Machine-learning-assisted descriptor identification for catalysts in catalytic ozonation of VOCs

Abstract

The catalytic ozonation method is a low-energy degradation technology for volatile organic compounds (VOCs). Developing efficient catalysts is crucial for advancing this technology. Up to 100 relevant works have been published to date, but analyzing the collective inferences through conventional literature searches is a challenging task. A machine learning (ML) framework was presented to create a database for catalytic ozonation of VOCs and to predict catalyst performance from experimental descriptors. Notably, the data mining process collected 577 data points with 21 descriptors. The VOC conversion rate was successfully predicted by the GBDT. Under constrained conditions, Shapley Additive exPlanations (SHAP) analysis revealed that Fe/MnO_x and Ce/MnO_x are the most promising catalysts for the catalytic ozonation. Two series of catalysts were prepared using different methods and evaluated for the catalytic ozonation of p-xylene and 1,2-dichloroethane (1,2-DCE) under constrained conditions, which validated the accuracy of the SHAP analysis. This work highlights ML as a practical tool for efficient design of catalysts, and provides appropriate descriptors for ML-assisted development of catalysts for catalytic ozonation of VOCs. The protocol allows for the prediction of promising promoter combinations for catalysts under various constrained conditions, providing guidance for catalyst design and development.