Water Pump Operation Optimization under Dynamic Market and Consumer Behaviour

In the face of growing energy and water consumption, the pumping costs of water supply systems in high-rise buildings are on the rise. The state of practice uses statically configured water level thresholds or time-based triggers to activate water pumps, while state-of-the-art research works propose to minimize pumping costs by dynamically adjusting the pump schedules.However, the implications of volatile energy price, dynamic consumer water demands, and other important factors - in particular, the impact on water pump health and the disturbance to residents by activating pumps during the night - have not been thoroughly considered in those research works.There is also a lack of thorough evaluation of their performance using real-world data over a prolonged period.Our work addresses those gaps by introducing a model predictive control optimization framework that incorporates machine learning predictions to handle water demand and energy price uncertainty. It combines multiple factors including pump health and resident satisfaction level to find an optimal solution.We used real-world data over prolonged periods of time that exhibit significant pattern changes to evaluate the performance of our dynamic scheduling solution. While significant gain is achieved over state-of-the-art and state-of-the-practice solutions, we also observed considerable amount of fluctuation in performance gains of such dynamic schemes, especially under varying prediction accuracy of water demand and energy price forecasting, which calls for more future research.