Flexibility Prediction in Wastewater-Energy Nexus using Machine Learning

Electric motors in applications such as pumps, fans and compressors are responsible for over half of the electricity consumption in developed countries. Some of these motors have a certain flexibility that can be aggregated and used to stabilise the electrical network. Flexibility is the duration that a motor can be altered without impacting functional requirements. The pump stations can unlock their flexibility through adaptive control of power-electronic-based variable speed drives. However, the application’s availability must be correctly predicted so that the critical applications controlled by power electronic drives are not compromised when operated outside standard conditions (i.e., to provide flexibility). This work identifies short- and long-term flexibility prediction methods for wastewater pumps. First, a method is developed to clean up and improve the real-world data, which is then used to develop two different models. The first model predicts how long a pump can be turned off without endangering its application. The second model performs a 24-hour-ahead prediction for the expected load on the pump. We discuss strengths and weaknesses and compare both models, identifying the best way to arrange both models to achieve an improved flexibility prediction.