Machine learning reveals structure-performance relationships of dry reforming of methane catalysts and the potential influencing mechanisms

Abstract

Catalyst design is crucial for optimizing dry reforming of methane (DRM), but traditional experimental design and computational methods are time and resource costly. Machine learning (ML) can help develop effective catalysts to some extent, we propose an interpretable ML model that achieves an R² of 0.91, and use tools to analyse the importance and interactions of the parameters involved in the reaction process. In addition to the well-known temperature and GHSV, the study reveals the potential effect of the calcination temperature on methane conversion, i.e., it affects methane conversion by influencing the surface structure of the catalyst. To further illustrate the model's ability to predict unknown variables, we chose variables that were not included in the dataset for experimental validation and were within 10 % of error. The reaction conditions were optimized for particular scenarios in the extended research. This method was successful in limiting the ideal reaction conditions to a particular range, which yielded fresh catalyst design concepts.