Data Science for Detecting the DO Elbow: Saving Energy During Aeration Cycles

Dissolved oxygen and redox potential sensors are the main tools used to control and monitor aerobic and anoxic cycles in the biological treatment tank of a wastewater treatment plant. Just running the aeration uses a lot of electricity. In this study, data analytics and machine learning were used to collect and understand the data and then model and analyse it to find the DO-Elbow, which can help to understand biological behavior and save energy in wastewater treatment. We used data from a cycle alternating between aeration and non-aeration phases to train a linear regression model on dissolved oxygen levels. The model achieved its lowest mean squared error of 0.9362 when the data was segmented into 20 to 100 segments. Consequently, four apparent DO-Elbows were observed during the aeration phase and one or two during non-aeration when 20 segments were used. These points indicate major changes in DO behaviour, typically associated with shifts in the biological or operational processes of the treatment system.