Performance Evaluation of Domestic Wastewater Treatment Using SBR With Air Recirculation and Optimize Its Performance by Machine Learning Approach

ABSTRACT

Conventional sequencing batch reactors (SBRs) suffer from poor oxygen utilization efficiency, typically consuming only 1%–2% of supplied oxygen while wasting 18%–20% in the exhaust gas. Although conventional intermittent aeration can lower energy costs, it frequently leads to rapid dissolved oxygen (DO) depletion during blower‐off periods, causing biological process failure. To address these critical limitations, this study presents a unique scientific contribution through the synergistic integration of a physical air‐recirculation system with a predictive machine learning (ML) optimization framework. A laboratory‐scale SBR equipped with a pressurized tank was designed to capture and reuse aeration off‐gas, improving oxygen utilization and reducing blower operating time. Continuous aeration achieved optimal performance at 4 h, with COD and NH ₄ ⁺ removal efficiencies of 72.1% and 91.7%, respectively, but energy demand increased proportionally with aeration duration. In contrast, intermittent aeration with recirculation maintained comparable pollutant removal while reducing energy consumption by 30%–40%. The 5 min on/5 min off regime provided the best balance, sustaining dissolved oxygen levels and stable biological activity. Machine learning models further enhanced optimization: Support vector regression (SVR) accurately predicted energy use, while XGBoost captured nonlinear pollutant removal dynamics. Model‐based response surfaces identified moderate intermittent aeration (4‐ to 6‐min cycles, 4‐h duration) as the most favorable operating region. This integrated approach offers strong potential for sustainable decentralized wastewater treatment.