Rapid Prediction of Average Intercalation Potential and Formation Energy of Decoupling Water-Splitting Buffer Electrode Materials Based on Machine Learning

Ion battery materials are applied as decoupled water-splitting buffer electrodes, enabling the temporal and spatial separation of hydrogen and oxygen production. However, developing materials based on experimental trial-and-error methods has been time-consuming and resource-intensive. Given that the average potential and average intercalation potential of ion battery materials contribute to the selection of solid-state redox mediators for decoupled water-splitting, we developed an integrated method combining high-throughput first-principles calculations and deep learning models to efficiently and accurately predict materials suitable for decoupled water-splitting buffer electrodes. A data set containing intercalation potential and formation energy entries, computed using density functional theory (DFT), was created. Machine learning models were trained on this data set, and the most influential features were identified through relevant interpretability frameworks. The analysis revealed that the mean electronegativity of elements was a key factor affecting the average intercalation potential, while the range of electronegativity exhibited a more complex influence on the formation energy. Using the developed model, the time required to determine the average intercalation potential was reduced to 1000th of that needed by traditional DFT methods. These findings highlight critical factors influencing material properties, offering valuable guidance for material design and significantly accelerating the initial screening process.