Machine learning enabled carbon-contaminant synergistic control in wastewater treatment plants toward carbon neutrality.

Wastewater treatment plants (WWTPs) are essential infrastructure for safeguarding water environmental security, while also serving as significant sources of greenhouse gas emissions. Given the dual constraints of carbon mitigation and contaminant removal, making dynamic decisions on operational parameters is challenged by the complex effects of multiple variables. Herein, carbon-contaminant synergy indices were proposed to evaluate the integrated performance in carbon mitigation and contaminant removal across three full-scale WWTPs, with machine learning (ML) models developed from long-term operational monitoring data to predict these indices. The developed ML models demonstrated robust predictive performance across all three WWTPs, with test R2 values ranging from 0.76 to 0.85. Shapley additive explanations (SHAP) analysis and classification mapping revealed that operational parameters exerted non-negligible influence on model outputs, accounting for 14.6%-43.8% of the total average absolute SHAP value across all variables. Additionally, Bayesian optimization was used to precisely regulate the operational parameters, yielding improved carbon-contaminant integrated performance across all three WWTPs. Overall, a feasible and universal ML framework was developed for WWTPs to support carbon-contaminant synergistic control and guide dynamic decision-making on operational parameters.