Machine Learning Predictive Models, Interpretations, and Applications of Metal-Based Catalysts, UV, and Thermal Activation Persulfate-Based Advanced Oxidation Processes

초록

This study develops a machine learning (ML) framework to predict pollutant degradation kinetic constants (k) in persulfate-based advanced oxidation processes (PS-AOP) activated by metal-based catalysts (MBC), UV, and thermal energy. Three data sets were constructed based on activation mechanisms, pollutant properties, and reaction conditions. After systematic data preprocessing, five ML algorithms were trained with 10-fold cross-validation and hyperparameter optimization to enhance accuracy and prevent overfitting. The support vector regression (SVR) model achieved the best performance for the MBC data set (R2= 0.88, 439 data points), while the multilayer perceptron (MLP) model excelled for UV (R2= 0.93, 273 data points) and thermal activation (R2= 0.94, 241 data points). The feature importance analysis was performed for revealing the data pattern and feature interpretation by using SHapley additive explanations (SHAP). A new method was introduced to validate the predictive model through scenario simulations based on existing literature, and particle swarm optimization (PSO) was employed for inverse design to optimize the reaction parameters. Additionally, electrical energy per order (E<inf>Eo</inf>) calculations were used to guide the selection of energy efficient activation strategies, considering both energy consumption and kinetic constants. The E<inf>Eo</inf>for the MBC, UV, and thermal systems were 3.06, 46.77, and 51.20 kW h m–3order–1, respectively, indicating that the MBC system was the most energy efficient. This study provides a robust predictive and interpretative tool for PS-AOP, aiding in practical implementation and optimization of pollutant degradation processes. © 2025 American Chemical Society

키워드