Unveiling the Mechanisms of Catalytic CO2 Electroreduction through Machine Learning

Abstract

The discovery and optimization of electrocatalysts used in the electro-reduction reaction of CO2 (CO2RR) to achieve high activity and selectivity is a costly and time-consuming process. Due to environmental concerns and the pivotal role of these catalysts in curbing the escalating consumption of fossil fuels, it is imperative to explore alternative methods for discovering electrocatalysts with superior performance in CO2RR. In this context, the application of machine learning (ML) to a comprehensive data set derived from experimental articles on electrocatalysts used in CO2RR is proposed, and the most influential parameters of highly promising catalysts for CO2RR were optimized. The catalyst exhibiting the highest faradaic efficiency (FE) of 95-100% in electrochemically producing CO is characterized by the following properties: metal content ranging from 2.5 to 7.5%, metal-N content ranging from 1.5 to 2.5%, total N content ranging from 2.0 to 7%, metal-N bond length ranging from 1.35 to 1.55 angstrom, free-energy barrier for *COOH ranging from -0.25 to 0.75 eV, free-energy barrier for *CO ranging from -1.5 to -0.25 eV, pore size between 7.0 and 15 nm, and a surface area of the carbon support within the range of 350-700 m(2)/g. The optimal potential is determined between -1.0 and 0.0 V versus a reversible hydrogen electrode, with a predicted stability of over 80 h. These findings demonstrate the potential of ML models, especially for a limited amount of experimental data, to provide desirable predictions for the design of more efficient electrocatalysts for CO2RR.