Error estimation in POD-based dynamic reduced-order thermal modeling of data centers

Abstract

A proper orthogonal decomposition (POD)-based reduced-order modeling framework that predicts transient air temperatures in an air-cooled data center is developed. The framework is applied on an initial temperature data set acquired by measurements at discrete time instants. The subsequent data analysis predicts air temperatures for times both inside and outside of the discrete time domain. The predicted temperature data are compared with corresponding experimental observations, and the prediction error is analyzed. An alternative analytical approach is developed for determining the error, and an iteration-based optimization procedure is developed to calibrate the analytical error against the POD-based modeling error. The calibrated analytical error is added to the corresponding POD-predicted temperature data to obtain reliable new temperature data.