Integrated optimization of effluent quality, energy use, and chemical dosing in wastewater treatment using a cyclic...

Wastewater treatment plants (WWTPs) still face challenges in simultaneously achieving stable effluent quality, low chemical dosing, and low energy consumption under dynamic influent conditions. Yet conventional controls often chase these disturbances after-the-fact, leading to excessive aeration, PAC overdosing, and repeated effluent breaches. To overcome these limitations, this study proposes a Cyclic Residual-Steered Multi-Objective Bayesian Optimization (CRS-MOBO) framework. It couples residual cycle models - learned 24-h periodic influent patterns - with (i) a residual-guided preference-steering mechanism and (ii) a Bayesian multi-objective optimizer, so that dissolved oxygen set-points and PAC dosing are adjusted proactively. We benchmarked CRS-MOBO on the ASM-based SUMO (Dynamita) simulator feeding with real operating data from the Hongxing Pharmaceutical Industrial Park WWTP. Against conventional PID control and knowledge-based NSGA-II, CRS-MOBO reduced aeration energy consumption by about 3.1% and 1.9%, respectively, and lowered PAC chemical consumption costs by about 20.2% and 22.9%, while keeping all effluent quality indicators within regulatory discharge limits. Even when hit by non-periodic disturbances, CRS-MOBO continued to optimize stably and effectively.