Interpretable modeling for effluent organic load control and optimization in wastewater treatment

Effluent organic load (ECOD) prediction is crucial for optimizing wastewater treatment plant (WWTP) operations, particularly in controlling organic pollution and improving treatment efficiency. This study proposes a machine learning framework combining extreme gradient boosting (XGBoost) with Shapley additive explanations (SHAP) to interpret and predict ECOD in a full-scale WWTP. The XGBoost model shows strong performance with an R² of 0.919 and an RMSE of 0.791 mg/L. SHAP analysis identifies key variables, such as aeration rate, influent flow rate, and suspended solids removal rate, that significantly impact ECOD prediction. The partial dependence plots reveal nonlinear effects and interactions among key variables. These analyses uncovered operational thresholds and synergistic or antagonistic relationships, guiding actionable strategies for aeration control, process sustainability, and nutrient removal. This framework enables adaptive operational adjustments, substantial energy savings, and emission reductions, providing a scalable and practical solution for enhancing process efficiency and environmental sustainability in WWTPs.