A Methodology to Develop Monthly Energy Use Models From Utility Billing Data for Seasonally Scheduled Buildings: Application to Schools

Abstract

The measured energy savings from retrofits in buildings in the Texas LoanSTAR program are determined as the difference between the energy consumption predicted by a baseline model and the measured energy consumption during the post-retrofit period. Savings measurement for buildings such as primary and secondary schools is very difficult due to the special operating schedules of these buildings. Currently, savings are often determined by simple pre-post utility bill comparison; they may also be determined with two separate models for the baseline: a 3-P model for non-summer months, and a mean model for the summer months. (Landman 1996). This thesis proposes a methodology for developing baseline models of energy use for buildings such as schools which have important daily and seasonal variations in occupancy. The method utilizes utility billing data, but also explicitly incorporates occupancy rate, permitting a generalized model which retains the distinction between energy use levels during occupied and unoccupied days of the year. The proposed methodology has been evaluated against the one proposed by Landman for 10 schools in Texas. The major results are summarized below: 1. The CV (Coefficient of Variation of the Root Mean Square Error ) values for the proposed methodology are much smaller than those of the 3-P mean model method, while the average absolute percent error is somewhat smaller for the proposed method, implying that it is suitable for developing baseline models for buildings such as schools that experience large seasonal changes in occupancy patterns. Although this method is a little more complicated it allows a more intuitive and unified model to be identified than the standard 3-P model. 2. Using daily data from the Dunbar Middle School, it is illustrated that the effect of the schedule on energy use is sometimes comparable to that of outside temperatures for heavily scheduled buildings. This suggests that selection of data periods for baseline model identification should be done with great care. 3. The proposed 4-P multiple-linear regression model is recommended. It was found to be somewhat more accurate than the 3-P mean model approach recommended by Landman.