Feature-engineered machine learning for daily-scale prediction of effluent total phosphorus and coagulant dosing...

Area of Science:

• Environmental Engineering
• Water Treatment Technologies
• Machine Learning Applications

Background:

• Effective phosphorus control is crucial for wastewater treatment plants to meet regulatory compliance.
• Dissolved air flotation (DAF) systems require precise operational decisions for phosphorus removal.
• Limited online sensing restricts DAF operational adjustments to daily time scales.

Purpose of the Study:

• To develop an interpretable, feature-engineered machine learning framework for daily prediction of effluent total phosphorus (T-P).
• To optimize coagulant dosing in a full-scale municipal DAF system using sensitivity analysis.
• To demonstrate the practical value of daily-scale machine learning for data-driven DAF operation.

Main Methods:

• Preprocessing of long-term operational, water-quality, and meteorological data (1,096 days) using outlier screening and imputation.
• Incorporation of mechanistically informed features including influent loading, operational conditions, and short-term T-P variability.
• Evaluation of machine learning models, with Random Forest achieving the best performance (Test R² = 0.818).

Main Results:

• Random Forest model predicted effluent T-P with an RMSE of 0.032 mg/L, within 20% of the 0.2 mg/L discharge limit.
• SHAP analysis identified influent T-P, coagulant dosage, and short-term variation as key drivers.
• Optimized dosing reduced coagulant consumption by 32-51%, with estimated annual savings of 1.53 billion KRW.

Conclusions:

• The proposed machine learning framework enables accurate daily prediction of effluent T-P in DAF systems.
• Sensitivity-based optimization significantly reduces coagulant usage and operational costs.
• Interpretable machine learning offers a practical approach for enhancing DAF system performance and compliance.