An AI-enabled optimal control strategy utilizing dual-horizon load predictions for large building cooling systems and its cloud-based implementation

Central cooling systems with multiple chillers are commonly used in large buildings due to their flexibility and efficiency in providing cooling capacity. These systems often rely on load prediction to optimize operations, ensuring that the cooling output aligns with immediate demand and also accounts for anticipated changes in future cooling demand. However, conventional prediction-based optimal control strategies usually utilize single-horizon prediction, typically either hour-ahead or tens-of-minutes-ahead load predictions, which cannot accurately capture both the short-term (minutes and hours) fluctuations and daily operation patterns of the complex cooling systems. This study proposes to utilize load predictions with dual prediction horizons for different optimal control tasks of large central cooling systems, including morning startup, chiller sequencing, and the chilled water supply temperature setpoint reset. Additionally, a method based on ensemble learning and Automatic Machine learning (AutoML) is proposed to generate prediction models. Initially, the day-ahead scheduling for the operation sequence of multiple chillers and the morning startup time is determined based on the probabilistic day-ahead cooling load predictions. Subsequently, near real-time adjustments to the chilled water temperature setpoint and chiller sequencing are executed based on an hour-ahead load prediction. The effectiveness of the proposed strategy is validated in a high-rise office building in Hong Kong through a cloud-based platform. The on-site test shows the average energy saving is 18.4%. The strategy can work on top of the existing building management systems and be conveniently deployed to many buildings, facilitating the building sector to realize energy saving and carbon reduction.