Machine learning-based optimization of enhanced nitrogen removal in a full-scale urban wastewater treatment plant...

Urban wastewater treatment plants (WWTPs) employing ecological combination ponds (ECPs) in small towns across China are struggling to dynamically adjust operating parameters under fluctuating influent quality conditions, resulting in excessive aeration and external carbon source overdosing to meet the stringent total nitrogen (TN) discharge standards. To address these issues, it is essential to establish an appropriate enhanced TN removal model for WWTPs with ECPs. In this study, we collected three years of operational data from a full-scale urban WWTP utilizing ECPs and implemented an interpretable machine learning approach to predict and optimize effluent TN concentration. The XGBoost model attained R² values of 0.997 and 0.911, RMSE values of 0.196 and 1.283 for the training and testing sets, respectively. The Shapley additive explanation analysis and partial dependence plots identified optimal operating parameters to improve TN removal while balancing energy consumption and chemical oxygen demand (COD) dosage reduction. A graphical user interface was developed to facilitate ongoing prediction and coordinated optimization of process operational parameters, achieving simultaneous reductions in effluent TN, energy consumption, and external carbon source usage. Notably, the effluent TN concentration decreased by 17.50 %, while COD dosage was reduced by 33.29 % annually. Consequently, WWTP with ECPs demonstrated substantial potential for carbon emission reduction. Total annual carbon emission reductions (788.40 t CO₂/y) were calculated solely based on enhanced TN removal, along with reductions in energy consumption and COD dosage. Our findings provide the optimal model for enhanced TN removal in urban WWTPs utilizing ECPs under variable influent quality pressure, thereby meeting stringent TN discharge standards and contributing to energy savings and carbon emission reduction goals.