Data-driven design of dual-metal-site catalysts for the electrochemical carbon dioxide reduction reaction

Electroreduction of carbon dioxide (CO2) offers a sustainable approach to realize carbon recycling and energy regeneration, and development of high-efficiency electrocatalysts for the CO2 reduction reaction (CO2RR) is the key scientific issue. Presently, dual-metal-site catalysts (DMSCs) have shown large potential in the electrochemical CO2RR; however, regulating the combination and structure of many diverse transition metals is a huge challenge. Herein, we created a rational machine-learning (ML) approach to investigate the reaction activity and selectivity of 1218 DMSCs toward CO2 electrochemical reduction. The gradient boosting regression (GBR) model possessing 17 features exhibited the best prediction accuracy with a root-mean-square error of 0.09 V and a coefficient of determination value of 0.98. By implementing two rounds of rigorous feature selection process, the screening model successfully predicted 4 DMSCs (Mn-Ru, Mn-Os, Zn-Ru and Co-Au-N-6-Gra-model 3) identified as efficient CO2RR electrocatalysts, and then these data of ML prediction were verified by density functional theory (DFT) calculations with high accuracy (less than 0.07 V error). This work demonstrates the immense potential of ML methods and provides an efficient and accurate screening approach for the rational design of high-performance electrocatalysts.