Adaptive data-driven optimization of chiller loading with domain knowledge

Abstract

In order to ensure energy saving and reliable operation for the actual multi-chiller systems, this paper proposes a novel chiller loading strategy that combines the operation data and domain knowledge. The operation data are used to establish the near-optimal performance map which can guide the optimization of chiller loading. The domain knowledge is used to formulate rules to keep the operation reliability of multi-chiller systems. Then a comprehensive method is developed to determine the near-optimal load allocation by considering the map, rules, and the current operation condition. Both the performance prediction study and on-site study are conducted to validate the proposed control strategy. In the performance prediction study, the strategy is compared with other two reference strategies during March to November. The results show that the proposed strategy has the highest energy efficiency among the three strategies. In field study, the proposed strategy is compared with the original strategy on the actual multi-chiller system. The on-site results show that the total energy saving of the proposed strategy is 5.3% compared with that of the original strategy. Then the proposed strategy can adapt to different operation conditions, ensuring energy savings while operating within the plant's constraints.