Data-Driven Development of Heterogeneous Catalysts for Propane Dehydrogenation with Machine Learning and Metaheuristic Optimization

Recent advances in data-driven approaches using the machine learning (ML) method have enabled the discovery of high-performance materials. This paper presents a hybrid framework that combines ML models with a metaheuristic optimization algorithm, to explore improved heterogeneous catalysts for propane dehydrogenation (PDH). The framework proposes multiple PDH catalysts, utilizing our laboratory-scale database. A unique five-component catalyst, 2.4Ga 2.2Pt 1.7B 1.3Zr/Al2O3, exhibits superior performance, achieving a propylene yield of 58% at 600 °C. This work highlights the excellent predictive capability of the framework and offers a new data-driven approach for developing high-performance materials for heterogeneous catalysis.