Machine Learning-Assisted Screening of Cu-Based Trimetallic Catalysts for Electrochemical Conversion of CO2 to CO

The electrochemical reduction of carbon dioxide to useful chemicals and fuels is a new strategy to utilize large amounts of carbon dioxide. However, the lack of efficient catalysts has hindered the development of this technology. Herein, a machine learning (ML)-assisted screening model is developed to explore efficient trimetallic electrocatalysts for the CO2 reduction reaction by combining with density functional theory (DFT) and electrochemical experiments. The group of doped elements in the periodic table is the most important descriptor of Cu-based trimetallic electrocatalysts and the support vector regression algorithm has the best predictive performance. Based on ML predictions, the overpotential of PdPt@Cu is successfully predicted to be 0.11 V, and it shows the best electrocatalytic performance for the CO2 reduction reaction (CO2RR). DFT calculation results show that CO2 → COOH* is the potential-limiting step of CO2RR-to-CO for PdPt@Cu and its overpotential is 0.09 V, which is consistent with the ML-predicted results. The electrochemical experiments show that the Faraday efficiency of CO is 82.12% at −0.8 V vs RHE for PdPt@Cu. After 12 h of electrolysis in the H-cell, the catalyst still maintains good catalytic performance. This work provides an efficient method for screening catalysts.