Machine learning-optimized advanced oxidation for enhanced sludge dewatering: EPS mechanistic insights and predictive modeling

The recalcitrant nature of extracellular polymeric substances (EPS) in sewage sludge severely limits dewatering efficiency. While advanced oxidation processes (AOPs) disrupt EPS matrices, their optimization remains challenging. This study integrates machine learning (ML) with AOPs to establish predictive frameworks for parameter optimization. A Bayesian-optimized XGBoost model (test R² = 0.87, based on a 70/30 train-test split) outperformed other algorithms in predicting optimal AOP configurations, while an AdaBoost-based model (test R² = 0.81) provided mechanistic insights. Radical donor and catalyst concentrations exhibited synergistic effects (r > 0.8) in hydroxyl radical generation, with pH and VS/TS ratio critically influencing EPS dynamics. Soluble EPS (S-EPS) dominated dewaterability control, whereas tightly bound EPS showed negligible impact. SHAP (SHapley Additive exPlanations) analysis identified radical donor dosage, catalyst loading, and pH as pivotal operational parameters, with acidic conditions enhancing EPS disruption. This work advances data-driven AOP optimization for sludge management, highlighting the need for dynamic EPS transformation studies and adaptive control systems to achieve sustainable wastewater treatment.