Bridging Dissolved Organic Matter Reactivity to Ozonation Catalysts for Cu@Al2O3 from the Molecular Level by Machine Learning

Catalytic ozonation is a widely used advanced oxidation process for treating refractory organic wastewater; yet, the variability in dissolved organic matter (DOM) composition complicates reaction mechanisms. A critical challenge lies in designing optimal catalysts tailored to wastewater characteristics, and this factor has seldom been systematically explored. Here, we integrated principal component analysis with correlation analysis to link wastewater properties to catalyst structural descriptors. Representative catalyst Cu@Al2O3 was used to treat three refractory wastewaters via catalytic ozonation, revealing stark differences in total organic carbon removal efficiency (19.1%–58.6%). Fourier transform-ion cyclotron resonance-mass spectrometry uncovered molecular-level heterogeneity in refractory organics, while a random forest model classified removed, resistant, and produced molecules with accuracies of 67.3%–80.4%. Removed molecules were predominantly aromatic, heteroatom-rich (N, S), and high molecular weight (>400 Da). Statistical modeling identified the indicator UV absorbance at 254 nm (UV254) as a robust surrogate for wastewater characterization. Mechanistically, the oxygen vacancy concentration strongly correlated with CHOS compound removal (r = 0.998), while hindered the degradation of fluorescence region V components (r < −0.997). This study demonstrates a data-driven strategy of bridging molecular DOM profiling and catalyst descriptors, to guide the rational design of ozonation catalysts for targeted wastewater treatment.