Machine Learning-Assisted Expert Control in Wastewater Aeration Treatment Process Application

Wastewater treatment plants (WWTPs) exhibit significant variability in process design and operation, therefore limiting the generalizability of mechanistic models for process control. Leveraging advances in the Industrial Internet of Things (IIoT) and artificial intelligence (AI), this study develops, implements, and validates five aeration control strategies in a full-scale WWTP: a conventional method (PID), two standalone models: long short-term memory (LSTM) and transformer and two hybrid models that integrate LSTM-based dissolved oxygen (DO) prediction with either a Decision Tree or a Fuzzy Inference controller. Novelly, we proposed evaluation criteria tailored to real-world WWTP operational goals and constraints. Among our control strategies, the LSTM–Fuzzy Inference hybrid model demonstrates the most robust performance, maintaining DO levels within ±0.5 mg/L of the set point for 53 consecutive days with almost no operator intervention. Over a 6-month deployment, it reduced DO overshoot events by over 90%, stabilized effluent quality within limits, and lowered energy use from the previous baseline. Operational expenditure (OPEX) decreased by approximately $180,000 compared to historical data. Our findings highlight the effectiveness of combining machine learning with expert control logic to enhance process stability and energy efficiency in WWTPs, providing a viable pathway for scalable, intelligent automation in wastewater treatment systems.