Machine Learning-Driven Global Optimization of Single-Atom Catalyst-Mediated Advanced Oxidation Processes

Single-atom catalysts (SACs) are state-of-the-art for advanced oxidation processes (AOPs) for purifying water contaminants. While previous studies have explored individual influencing factors, such as the central metal species and coordination environment, reaction conditions, or contaminant molecular properties, the combined effects of these variables on AOP kinetics and thermodynamics remain poorly understood. Here, we propose a machine learning model based on a global optimization strategy that leverages a random forest model to predict pollutant degradation performance with high accuracy. The d electron number of the central metal and the average electronegativity of the coordination environment are identified as key descriptors in determining AOP performances. Theoretical calculations, including charge density distribution, adsorption energy, projected density of states, and crystal orbital Hamilton population metrics, reveal strong linear relationships between these descriptors and peroxymonosulfate activation energy. Global optimization analysis reveals that the optimal catalyst configuration requires metals possessing 5–7 d electrons, combined with coordination environments with average electronegativity values below 3.04. In addition, contaminant characteristics significantly affect degradation performances. Specifically, faster pollutant degradation is realized for organics with energy gaps below 3.92 eV and dipole moments greater than 7 D. This study offers a machine learning-guided pathway for intelligent design of SACs for effective AOP-based purification systems.