Physically consistent data-driven optimal sequencing strategy for variable speed pumps in large building chiller plants

Abstract

Variable speed pumps (VSPs) are widely adopted in HVAC systems for delivering chilled water to reduce energy consumption under partial load conditions. However, in large chiller plants with multiple parallel VSPs, pump sequencing is often rule-based without further optimization, failing to achieve optimal energy efficiency. Furthermore, previous optimization methods often rely on manufacturers’ curves, which lack reliability for practical implementation. Therefore, this study proposes a physically consistent data-driven optimal sequencing strategy to minimize energy consumption for parallel VSPs. The strategy involves two core components: (1) Interpretable power models trained on historical data predict total power consumption based on operating speed and total flow rate for each potential number of operating VSPs. (2) A physically consistent prediction method predicts the required operating frequency for alternative VSP numbers while maintaining system conditions. The optimal number of VSPs is then determined based on the minimum total power. The proposed strategy was validated through data experiments and field tests in an educational building. The data experiments show that the proposed strategy has a 10 % annual energy-saving potential, and the four-day field tests reveal a 15 % energy savings compared to the rule-based baseline.