Machine Learning-Assisted Study of RENC-Doped Graphene as Potential Electrocatalysts for Oxygen Electrode Reactions.

In the application of renewable energy, the oxidation-reduction reaction (ORR) and oxygen evolution reaction (OER) are two crucial reactions. Single-atom catalysts (SACs) based on metal-doped graphene have been widely employed due to their high activity and high atom utilization efficiency. However, the catalytic activity is significantly influenced by different metals and local coordination, making it challenging to efficiently screen through either experimental or density functional theory (DFT) calculations. To address this issue, this study employed a combination of DFT calculations and machine learning (DFT-ML) to investigate rare earth-modified carbon-based (RENC) electrocatalysts. Based on computational data from 75 catalysts, we trained two ML models to capture the underlying patterns of physical properties and overpotential. Subsequently, the candidate catalysts were screened, leading to the discovery of four ORR catalysts, nine OER catalysts, and five bifunctional electrocatalysts, all of which were thoroughly validated for their stability. Lastly, by integrating the ML models with the SHAP analysis framework, we revealed the influence of atomic radius, Pauling electronegativity, and other features on the catalytic activity. Additionally, we analyzed the physicochemical properties of potential catalysts through DFT calculations. The revolutionary DFT-ML approach provides a crucial driving force for the design and synthesis of potential catalysts in subsequent studies.