Reduced-order modeling framework for improving spatial resolution of data center transient air temperatures

Abstract

A proper orthogonal decomposition (POD)-based modeling framework is developed for improving the spatial resolution of transient rack air temperature data collected in a heterogeneous data center (DC) facility. Blocking cooling air inflow into racks periodically, three sets of transient temperature data are collected at the outlets of electronic equipment residing in three different racks. Using various combinations of initial discrete data as ensembles, the capability of the proposed POD/ interpolation framework for predicting new temperature data is demonstrated. The accuracy of POD-based temperature predictions is validated by comparing it to corresponding experimental data. The root mean square deviations between experimental data and POD-based predictions are found to be on the order of 5%.