Prediction and optimization of in-situ excess sludge reduction in wastewater treatment process by machine learning

In-situ excess sludge reduction includes three main techniques: uncoupling metabolism, lysis-cryptic growth, and microbial predation. In this study, machine learning was utilized to analyze in-situ sludge reduction, addressing the gap between large dataset and industrial needs. The best prediction model was Random Forest (R² = 0.8). SHapley Additive exPlanation analysis showed that treatment dosage, type of reagent, sludge retention time, and treatment energy were important features for in-situ sludge reduction. Considering both sludge reduction efficiency and effluent quality, partial dependence plot (PDP) analysis showed that lysis-cryptic growth was the best in-situ sludge reduction techinique; ultrasound and ultrasound-uncoupling were the best sludge lysis methods. 2D-PDP showed that suitable conditions for lysis-cryptic growth were: sludge recirculation ratio 15-100 %, energy 0.36-1.8 W/mL, time 0-15 min, dosage 4-5 mg/L. This study provides a scientific foundation for the optimized control of in-situ sludge reduction processes, providing a reference for practical applications.