Lea Ga\u0161pari\u010d, Anton Kokalj and Sa\u0161o D\u017eeroski<\/p>\n
Abstract
The growing interest in hydrogen gas as a fuel drives research into environmentally friendly hydrogen production methods. One viable approach of obtaining hydrogen is the electrocatalysis of water, which includes the hydrogen evolution reaction (HER) as one of the half-reactions. In the search of highly active catalysts for the HER, machin learning can be effectively utilized to develop models for calculating hydrogen adsorption energy, a key descriptor of catalytic activity. In this study, we learned models for predicting hydrogen adsorption energy on platinum. We used various machine-learning (ML) techniques on two datasets, one for extended surfaces and the other for nanoparticles.
The respective results reveal that ML models for extended surfaces are more accurate than those for nanoparticles,and that the features describing the local environment are the most significant for the predictions. For surfaces, the coordination number is the most relevant feature, while the d-band center is the most important for nanoparticles. The ML models developed in this study lack sufficient accuracy to provide reliable results, highlighting the need for further investigation with additional features or larger datasets.<\/p>\n\n\n\n