Screening of steam-reforming catalysts using unsupervised machine learning

Abstract

An efficient catalyst is crucial for the production of hydrogen through methanol reforming. However, the current study of catalysts for this reaction primarily relies on time-consuming and labor-intensive experimental methods. In this article, we propose a novel screening strategy called the Parameter-free Multi-view Clustering via Interactive Graph Mining (PMC–IGM) model to identify highly active methanol-reforming catalysts. Our approach involves the design of a new element representation method and clustering model, which effectively integrate and utilize information from different aspects of elements. The model achieved a predictive accuracy (ACC) of 61% and a Rand index (RI) of 97%. Notably, the model successfully identified two methanol-reforming catalysts with promising results, validating the efficiency and rationality of the algorithm. This research holds great significance in accelerating the discovery of novel methanol-reforming catalysts.