334 Energy Project Financing: Resources and Strategies for Success
objective is to consider all factors creating uncertainty, either qualita-
tively or quantitatively.
The accuracy of a savings estimate can be improved in two general
ways. One is by reducing biases, by using better information or by using
measured values in place of assumed or stipulated values. The second
way is by reducing random errors, either by increasing the sample sizes,
using a more efficient sample design or applying better measurement
techniques. In most cases, improving accuracy by any of these means
increases M&V cost. Such extra cost should be justified by the value of
the improved information (see Chapter 4.11).
Quantified uncertainty should be expressed in a statistically mean-
ingful way, namely declaring both accuracy and confidence levels. For
example, “The quantifiable error is found, with 90% confidence, to be
+20%.” A statistical precision statement without a confidence level is
meaningless since accuracy can sound very good if the confidence level
is low.
The appropriate level of accuracy for any savings determination is
established by the concerned parties. Appendix B discusses some issues
in establishing a level of uncertainty.
For buildings, one or more full years of energy use and weather
data should be used to construct regression models. Shorter periods
introduce more uncertainty through not having data on all operating
modes. The best predictors of both cooling and heating annual energy
use are models from data sets with mean temperatures close to the
annual mean temperature. The range of variation of daily temperature
values in the data set seems to be of secondary importance. One month
data sets in spring and fall, when the above condition applies, can be
better predictors of annual energy use than five month data sets from
winter and summer.
The required length of the metering or monitoring period depends
on the type of ECM. If, for instance, the ECM affects a system that
operated according to a well-defined schedule under a constant load,
such as a constant-speed exhaust fan motor, the period required to
determine annual savings could be quite short. In this case, short-term
energy savings can be easily extrapolated to the entire year. However,
if the project’s energy use varies both across day and seasons, as with
air-conditioning equipment, a much longer metering or monitoring pe-
riod may be required to characterize the system. In this case, long-term
data are used to determine annual energy savings.