| Abstract | Reliable measurement and verification (M&V) of the energy savings associated with energy conservation measures (ECMs) is critical for financial settlement of utility incentives and energy performance contracts. With the proliferation of advanced metering infrastructure, there is increasing interest in using whole-building data for M&V, as a cost-effective alternative to sub-metering or installing equipment-specific data loggers. Commonly observed protocols provide a framework for regression-based M&V analysis of whole-building data. This paper leverages seven years of whole-building electricity data from four geometrically-identical multi-storey multi-tenant co-located office towers, which underwent a series of very similar ECMs, to explore the application and validity of straightforward, regression-based models. Results showed the models met the criteria for goodness of fit and statistical uncertainty (R2 above 0.75; coefficient of variation of the root-mean-square error, CV(RMSE) below 25%; normalized mean bias error, NMBE below 0.005%; standard error in the unstandardized regression coefficient estimating an ECM effect less than 50% of the regression coefficient itself). Further, these regression models were (in the large majority of cases) able to differentiate between multiple ECMs applied sequentially, and that, in some cases, effects as low as 5% were reported. Effect estimates of similar ECMs were similar across the buildings, but were generally substantially larger than the pre-retrofit engineering predictions of savings. ECM effect estimates using monthly and hourly data with equivalent regression formulations gave very similar estimates, differences in coefficients ranged between approximately 0–15%, with a mean of 7%. Estimates of ECM effects made using the full timespan of data were more sensitive, and generally larger, than estimates using only a single year of pre- and post-ECM data. However, visually identifying ECM effects in time-series data was challenging, which highlights the many sources of error in such an analysis. |
|---|
