Using Machine Learning to Forecast the Conductive Substrate-Supported Heteroatom-Doped Metal Compound Electrocatalysts for Hydrogen Evolution Reaction

Abstract

The heteroatom-doped metallic compounds supported on conductive substrates are excellent catalysts for the hydrogen evolution reaction (HER) thanks to their tunable properties, e.g., metallic and nonmetallic compositions, especially bimetallic active centers and their synergistic effect, as well as the tunable morphology and interaction between the active centers and substrate. Only the optimal combination between these adjustable properties and other external factors could endow the remarkable HER catalytic activity of the catalysts. Therefore, in this study, the machine learning (ML) database based on plenty of HER catalysts from publicly available data was conducted to train three different ML models, and the various features including electrolyte type, catalyst morphology, compositions (metallic and nonmetallic) and their ratios, additive, and substrate were analyzed to figure out their impacts on overpotential (OP) values to determine the outstanding HER catalysts. According to the feature importance and Spearman coefficient analysis, the optimal combination of metal elements and their ratio were determined to be Pt, Mo and 0.5, and the heteroatoms and substrate were determined to be nitrogen, sulfur, and nickel foam. Finally, the ML model predicts that the foam nickel-supported bimetallic catalyst composed of Pt and Mo 2 S 3 and codoped with nitrogen and sulfur (N, Sdoped Pt@Mo 2 S 3 ) exhibits the admirable HER catalytic performance in alkaline electrolytes with a pretty low OP value of 33 mV. The database-guided ML model provides an alternative for rapid screening and prediction of HER electrocatalysts.