Improved Operation Forecasting of Air Blowers in a Wastewater Treatment Plant via Machine Learning-based Models

Wastewater treatment processes are energy intensive, thus requiring improvement for energy efficiency. Recent literature shows efforts to improve aspects of wastewater treatment processes with machine learning models. This study focuses on one particular process stage, the aeration stage of wastewater treatment, where bacteria is used for aerobic digestion of organic matter. This bacteria requires oxygen to live, requiring the use of air blowers to maintain the water's oxygen content at a minimum level. This aeration process comprises a significant percentage of the wastewater treatment plant's energy usage; thus, any improvements to this process have a large impact on a plant's overall energy usage. Hence, the goal of this study is to predict the operation intensity of air blowers based on influent water parameters using five machine learning models. Many plants rely on operators using experience and best judgment to make control decisions regarding the air blowers' operation. Reliance on operator experience allows for human error, reducing the energy-efficiency of a wastewater plant. For this study, data from the Benchmark Simulation Model No.1, or BSM1, is used to train several machine learning models in predicting the short-term future operation of a plant's air blowers for a given influent load, and their effectiveness is evaluated relative to one another. With this comparative evaluation, a forecasting model may be found which creates energy savings by helping operators make better-informed control decisions. Based on the example case study results, the proposed approach shows promise for this application (e.g., the Random Forest regression model performed with an average adjusted
$R^{2}$
score of 0.8960).