Machine Learning-Assisted Design of Long-Term Industrial Robust Catalysts for Selective Oxidation of Alcohol to Carboxylic Acid

The long-term stability of catalysts remains a key constraint on the industrial application of alcohol selective oxidation to carboxylic acids, as it is challenged by numerous deactivation factors from both micro and macro levels. Herein, this issue was systematically investigated by using machine learning. A multidimensional feature database encompassing catalyst composition and physical parameters was constructed. Multiple regression models were evaluated, and the best-performing tree-based models (Extreme Gradient Boosting and Gradient Boosting Regression) were further interpreted by SHAP, which revealed leaching resistance as the most significant contributor to stability, followed by mechanical strength and thermal stability. Accordingly, improvement strategies are proposed: constructing enhanced single atom metal–support interaction architectures to inhibit leaching; introducing binders during molding to improve wear resistance; and doping carbon materials to optimize thermal management. Based on the aforementioned insights, we successfully prepared Pt-s/HAP and ZnO/HAP catalysts using spray forming technology. In glycerol oxidation and isooctanol oxidation reactions, the two catalysts achieved stable operation for over 400 and 200 h, respectively, setting the current highest stability records for their respective catalytic systems. This successful case fully demonstrates the critical role of data-driven strategies in guiding the development and industrial application of high-performance catalysts.