Machine learning screening of high-performance single-atom electrocatalysts for two-electron oxygen reduction reaction

Electrocatalysis has emerged as one of the most promising alternatives to conventional anthraquinone for preparing hydrogen peroxide (H2O2) with high energy consumption and pollution because of its simplicity, convenience, and environmental friendliness. However, the oxygen reduction reaction (ORR) generating H2O2viathe2e- path is acompetitive path for 4e-ORR to generate H2O. Therefore, it is crucial to identify an electrocatalyst with high selectivity and activity of 2e-ORR. Here, we established five machine learning (ML) models based on the adsorption free energy of O* (△G (O*)) of 149 single-atom catalysts (SACs) collected and the limiting potential (UL) of 31 SACs calculated using density functional theory (DFT) from the literature. We then obtained descriptors that could accurately describe SACs. Furthermore, 690 unknown SACs' 2e-ORR catalytic performance was well predicted. Four 2e-ORR materials with high selectivity and activity were screened: Zn@Pc-N3C1, Au@Pd-N4, Au@Pd-N1C3, and Au@Py-N3C1. We verified the UL of these SACs through DFT calculation, which was higher than the standard value, proving the ML model's validity. The ML-based method to predict the material properties with highly selective and active electrocatalysts provides an efficient, rapid, and low-cost method for discovering and designing more valuable SACs catalysts.