Dynamic Optimization of a Liquid-Cooled Data Center Based on Hourly Outdoor Air Temperature and Humidity Using a Variable-Frequency Water Pump

Abstract

Current research primarily investigates the safe operating characteristics and energy-saving patterns of liquid-cooled data center chips under varying outdoor temperatures and constant humidity conditions. However, existing dynamic optimization methods overlook the problem of optimizing energy consumption while ensuring safe operation constraints during chip-load dynamic adjustments under the combined effects of dynamically changing temperature and humidity. While keeping servers secure and minimizing energy consumption, this study proposes a dynamic operation strategy that adopts a regression fitting model or a BPNN model combined with a variable-frequency pump for different chip loads based on dynamic weather conditions. The results show that compared with optimal dynamic operation considering only temperature and using a fixed-frequency pump, the proposed strategy using a variable-frequency pump reduces the secondary-side average flow rate by up to 6.77%, the primary-side average flow rate by up to 5.04%, and increases the average energy-saving rate by up to 18.15%. Additionally, with this variable-frequency pump method, dynamic optimization considering both temperature and humidity (rather than temperature alone) increases the primary flow rate by 101.09%, the secondary flow rate by 13.04%, and the energy consumption by 74.29%. It is also found that under variable outdoor temperature–humidity conditions, optimization with a variable-frequency pump differs significantly from optimization considering only dynamic outdoor temperature. Moreover, with this dynamic strategy (incorporating variable outdoor conditions and the variable-frequency pump), optimal operation varies greatly with different chip loads. This study offers an innovative method for the dynamic optimization of data center cooling under variable outdoor temperature and humidity, enabling energy savings while ensuring server safety.