Incremental machine learning and genetic algorithm for optimization and dynamic aeration control in wastewater treatment plants

Abstract

Wastewater treatment plants (WWTPs) play a crucial role in municipal infrastructure, but their energy consumption remains a significant concern. Among the various components of WWTPs, the aeration system in biological reactors stands out as a major contributor to high energy usage. This system accounts for >50 % of the plant's total power consumption, as it ensures the effective removal of organics and nitrogen. Supervisory Control and Data Acquisition (SCADA) systems are commonly employed to monitor dissolved oxygen (DO) concentration and regulate aeration blower to maintain a specific DO setpoint. However, despite the prevalence of SCADA systems, many WWTPs still grapple with challenges such as over-aeration and under-aeration caused by diurnal wastewater loading cycles, resulting in increased energy usage. To address this issue, this research introduces a predictive aeration optimization tool tailored to a full-scale biological nutrient removal WWTP. An incremental learning (IL) model based on K-Nearest Neighbor (KNN) that passively handles changing data patterns is developed to predict air blower flow rates, achieving an R² value that exceeds 85 %. This model further serves as an objective function for a Genetic Algorithm (GA) optimization, aimed at minimizing air blower flow rates while ensuring that final effluent properties meet treatment quality limits in compliance with regulatory requirements. The model is trained and validated using online sensor data collected from 2012 to 2022, with measurements taken every 10 min. When placed in a simulated production scenario, the model successfully optimized aeration requirements, achieving a 14 % reduction without compromising effluent quality.