Appendix B 323
restaurant sales.
To the extent that independent variables have a cyclical nature to
them, the significance of their impact on energy use can be assessed
through mathematical modeling. Parameters found to have a significant
effect in the baseyear period should be included in the routine adjust-
ments when applying equation (1) for determining savings. Parameters
having a less predictable but potentially significant effect should be
measured and recorded in the baseyear conditions and post-retrofit pe-
riods so that non-routine baseline adjustments can be made if needed
(see Chapter 4.8)
Independent variables should be measured and recorded at the
same time as the energy meters. For example, weather data should be
recorded daily so it can be totaled to correspond with the exact monthly
energy metering period which may be different from the calendar
month. Monthly mean temperature data for a non-calendar month
would introduce unnecessary error into the model.
The number of independent variables to consider in the model of the
baseyear data can be determined by regression analysis and other forms
of mathematical modeling (Rabl 1988, Rabl and Rialhe 1992, ASHRAE
1997, Fels 1986, Ruch and Claridge 1991, Claridge et al. 1994).
3.4.3.4 Option C: Data Analysis and Models
The adjustment term of equation (1) under Option C is calcu-
lated by developing a valid model of each meter’s baseyear energy
use and/or demand. A model may be as simple as an ordered list of
twelve actual baseyear monthly electrical demands without any adjust-
ment factors. However they can often be a set of factors derived from
regression analysis correlating energy use to one or more parameters
such as degree days, metering period length, occupancy, and building
operating mode (summer/winter). Models can also involve several sets
of regression parameters each valid over a defined range of conditions
such as ambient temperature, in the case of buildings, since buildings
often use energy differently in different seasons.
Option C usually requires 12, 24, or 36 (i.e., one full year or mul-
tiple years) of continuous baseyear daily or monthly energy data, and
continuous data during the post-retrofit period (Fels 1986) since models
with more or less data (i.e., 13, 14, 15 or 9,10, 11 months) can cause
the regression to have a statistical bias. Meter data can be hourly, daily
or monthly whole-building data. Hourly data should be aggregated at