Annual prediction accuracy of monthly regression models for energy consumption in commercial buildings - preliminary results
Conference Paper
Overview
Additional Document Info
View All
Overview
abstract
Regression models of measured energy use is commercial buildings are widely used as baseline models for determining retrofit savings from measured energy consumption. It is less expensive to determine savings from monthly utility bills when they are available than to install hourly metering equipment. However, little is known about the accuracy of savings determined from monthly data. This paper reports a preliminary investigation of this question by comparing the heating and cooling energy use predicted by regression models based on monthly data with the predictions of calibrated hourly simulation models when applied to a medium-sized university building in Texas with (i) DDCAV system operating 24 hours per day, (ii) DDCAV system with unoccupied shut down, (iii) DDVAV system operating 24 hours per day, and (iv) DDVAV system with unoccupied shut down. The results of the four cases studied indicate: 1) when the AHUs are operated 24 hours/day, the annual prediction error of the cooling regression models are less than 0.5% of the annual cooling energy consumption; however, 2) when the AHUs are operated with unoccupied shut down, the annual prediction error of the cooling models is as high as 6.1% of annual energy consumption. Modified regression models based on average temperature values during the operating periods are recommended when AHUs operated less than 24 hours/day and the temperature pattern of pre- and post-retrofit years are different. The modified cooling regression models reduced the annual prediction error to 0.6% from -6.1%. The modified heating regression model reduced the annual prediction error to 3.9% from 5.7%.
