Prediction and optimization of key factors for catalytic O3 degradation of antibiotics based on Catboost model coupled Bayesian optimisation algorithm

Abstract

Machine learning (ML) has the ability to improve the efficiency of heterogeneous catalytic ozone (O₃) degradation of antibiotics through the potential for rapid identification of key factors in the catalytic system. However, there is still a lack of research on the reverse regulation of catalyst metal composition and reaction conditions for specific antibiotics. This study combined ML model and the Bayesian Optimization Algorithm (BOA) to predict and optimize the key influencing factors in the catalytic O₃ degradation of antibiotic systems. The results indicate that the CatBoost model presented the optimal training efficiency and accuracy, with an R² value of 0.9482 and an RMSE of 0.0401. SHAP analysis revealed that antibiotic type, catalyst metal composition, and O₃ flow rate are the three most critical features in model training, with significant interactions observed between O₃ flow rate, pH, and catalyst dosage. Experimental validation demonstrated that the experimental TOC removal rate of antibiotics reached 87.33 % under optimized conditions, while the model-optimized TOC removal rate achieved 90 %, with an error of only 2.97 %. The present work significantly improved the efficiency and accuracy of O₃ catalytic degradation process design for antibiotics and provided a novel approach for selecting and predicting reaction parameters in O₃ catalytic degradation of specific antibiotics.