Rack Thermal Model Prediction Accuracy Improvement by Utilizing Rack Power Consumption as Modelling Feature

摘要

In this paper, importance of rack power consumption as modelling feature of data center thermal model is studied. Data center thermal model is utilized to predict rack inlet and outlet temperature distribution in order to control air conditioner operational parameters such as supply/return air temperature, air conditioner indoor unit fan speed etc. Experimental data collection has been performed at a data center in Southern India, consisting of 18 racks and 3 air conditioners. Rack temperature (inlet and outlet at top and middle position), rack power consumption and air conditioner return air temperature data was collected over 4 days. Return air temperature of air conditioner was varied at steps of 21°C, 23°C and 25°C on each day and total rack power consumption was found to be varying between 37.6 kW to 42.2 kW. Thermal model utilizing rack power consumption and air conditioner return air temperature as modelling features can predict rack inlet and outlet temperature with 0.64°C RMS error, showing an improvement of 26.4% over thermal model without rack power consumption and air conditioner return air temperature only as modelling feature. Also, error range reduces to 1.8°C, showing an improvement of 33.3%.