Model predictive control of large chiller plants for enhanced energy efficiency utilizing inherent cold storage of cooling systems

In many practical situations, chiller plants often operate continuously. However, during low-demand periods, such as night hours and the end of working hours, they often run inefficiently, leading to significant energy waste. This issue is especially challenging for constant-speed chillers. In fact, utilizing the inherent cold storage to “force” the chillers to operate at high loads and high efficiency is a practically attractive option. Two innovative chiller control strategies are proposed for night hours and the end of working hours, respectively, leveraging the inherent cold storage in chilled water distribution networks. These strategies employ a model-based approach using deep learning to enhance the chiller energy efficiency while maintaining acceptable start-stop frequency. Their effectiveness is limited to scenarios with low cooling loads and appropriate distribution networks. The strategies are implemented and tested in a real central cooling system of a large commercial building. Field test results show that the proposed control strategies can reduce total chiller power consumption by 28.1 % during the night hours and by 14 % at the end of working hours. • Two control strategies are proposed for utilizing system inherent cold storage. • The strategies significantly improve chiller efficiency under low partial load. • A deep learning method is developed for 1-h ahead building cooling load prediction. • The prediction method effectively reduces chiller switching in control systems. • On-site test results validate the applicability and effectiveness of the strategies.