Data-Driven Understanding of Dynamic Structural Restructuring in Perovskite Fluorides for Efficient Oxygen Evolution Catalysis

Electrochemical Oxygen Evolution Reaction (OER) is crucial for sustainable energy, but its sluggish kinetics demand efficient catalysts. Transition metal-based perovskite fluorides show promise due to dynamic structural restructuring (DSR) forming active oxyhydroxide layers. However, the microscopic mechanisms of DSR and its quantitative link to performance remain elusive with traditional methods. This study introduces a novel Hierarchical Ensemble Learning (HEAL) framework, integrating operando spectroscopy, electrochemical data, and theoretical calculations, to comprehensively understand DSR and optimize OER performance. The HEAL framework integrates three core modules: the Dynamic Restructuring Stage Identification Module (DRSIM) for accurate real-time DSR stage classification; the OER Activity Prediction and Optimization Module (OERPOM) for robust OER activity prediction; and the Microscopic Mechanism Interpretability Module (MMIM), leveraging Graph Neural Networks (GNNs) and SHAP value analysis to uncover critical physicochemical descriptors like F vacancy concentration and d-band center. Benchmarking against state-of-the-art models demonstrates HEAL's superior performance and interpretability. This data-driven approach offers unprecedented insights into complex electrocatalytic phenomena, providing a robust platform for rational design of high-performance catalysts.