Publication

Abstract : Fluctuations of Dissolved Oxygen (DO) are crucial to the biological activities in sewage treatment plants (STPs). It directly influences sewage treatment efficiency and energy consumption in the operation of STP. However, it is challenging to apply conventional methods to predict DO levels due to temperature variation and influent load fluctuations. This study investigates machine learning algorithms for forecasting minimum and maximum DO concentrations to help STP operators adjust the aeration system and process. The dataset consists of water quality features (e.g., temperature, pH, conductivity). This work trained and compared 11 machine learning models, including tree-based and boosting algorithms, to model the DO variations. Predicting minimum DO levels helps prevent anaerobic conditions and sludge bulking, while predicting maximum DO aids in minimizing energy waste due to over-aeration. This work built models forecasting both min DO and max DO. Key findings demonstrated that the CatBoost model performs best for forecasting both min DO and max DO across three metrics MSE (min DO: 0.2616, max DO: 0.2680), MAE (min DO: 0.3423, max DO: 0.3301) and R2 Score (min DO: 0.8888, max DO: 0.8196).