▎ 摘　　要

Single-atom catalysts are proven to be an effective strategy for suppressing shuttle effect at the source by accelerating the redox kinetics of intermediate polysulfides in lithium-sulfur (Li-S) batteries. However, only a few 3d transition metal single-atom catalysts (Ti, Fe, Co, Ni) are currently applied for sulfur reduction/oxidation reactions (SRR/SOR), which remains challenging for screening new efficient catalysts and understanding the relationship between structure-activity of catalysts. Herein, N-doped defective graphene (NG) supported 3d, 4d, and 5d transition metals are used as single-atom catalyst models to explore electrocatalytic SRR/SOR in Li-S batteries by using density functional theory calculations. The results show that M-1/NG (M-1 = Ru, Rh, Ir, Os) exhibits lower free energy change of rate-determining step (Delta GLi2S*)$( {\Delta {G}_{{\mathrm{Li}}_{\mathrm{2}}{{\mathrm{S}}}<^>{\mathrm{*}}\ }} )$ and Li2S decomposition energy barrier, which significantly enhance the SRR and SOR activity compared to other single-atom catalysts. Furthermore, the study accurately predicts the Delta GLi2S*$\Delta {G}_{{\mathrm{Li}}_{\mathrm{2}}{{\mathrm{S}}}<^>{\mathrm{*}}\ }$ by machine learning based on various descriptors and reveals the origin of the catalyst activity by analyzing the importance of the descriptors. This work provides great significance for understanding the relationships between the structure-activity of catalysts, and manifests that the employed machine learning approach is instructive for theoretical studies of single-atom catalytic reactions.