This  paper  documents  the  development  and  application  of  a  general  statistical  methodology to assess the accuracy of baseline energy models, focusing on its application  to  Measurement  and  Verification  (M&V)  of  whole-­‐building  energy  savings.  The methodology complements the principles addressed in resources such as ASHRAE Guideline  14  and  the  International  Performance  Measurement  and  Verification  Protocol. It requires fitting a baseline model to data from a ``training period’’ and using the  model  to  predict  total  electricity  consumption  during  a  subsequent  ``prediction  period.’’ We  illustrate  the  methodology  by  evaluating  five  baseline  models  using  data  from  29  buildings. The training period and prediction period were varied, and model predictions of  daily,  weekly,  and  monthly  energy  consumption  were  compared  to  meter  data  to  determine model accuracy. Several metrics were used to characterize the accuracy of the predictions, and in some cases the best-­‐performing model as judged by one metric was not the best performer when judged by another metric.

This paper documents the development and application of a general statistical methodology to assess the accuracy of baseline energy models, focusing on its application to Measurement and Verification (M&V) of whole-building energy savings. The methodology complements the principles addressed in resources such as ASHRAE Guideline 14 and the International Performance Measurement and Verification Protocol. It requires fitting a baseline model to data from a training period and using the model to predict total electricity consumption during a subsequent prediction period.We illustrate the methodology by evaluating five baseline models using data from 29 buildings. The training period and prediction period were varied, and model predictions of daily, weekly, and monthly energy consumption were compared to meter data to determine model accuracy. Several metrics were used to characterize the accuracy of the predictions, and in some cases the best performing model as judged by one metric was not the best performer when judged by another metric.

We analyze data from 36 commercial and government buildings that participated in a Demand Response (DR) program in California, to investigate the extent to which DR load shed in each building depends on outdoor air temperature, and whether the load shed varies systematically from year to year. Our baseline model has substantially lower error than other standard models but uncertainty in the load shed is still an impediment to addressing these questions. The model is accurate enough in 29 buildings to be used to investigate the relationship between outdoor temperature and the DR load shed, and data availability and accuracy are sufficient to investigate year-to-year persistence of load shed in 19 buildings. We find that for buildings in this dataset, most buildings shed several percent of their load during DR events. In about two thirds of buildings, higher outdoor air temperature lead to slightly reduced load shed. Year-to-year changes in load shed were generally small, except that in several buildings the load shed was small or nonexistent in the first year of participation in the program and increased subsequently.

With advances in energy metering, communication, and analytic software technologies,providers of Energy Management and Information Systems (EMIS) are opening new frontiers inbuilding energy efficiency. Through their engagement platforms and interfaces, EMIS productscan enable energy savings through multiple strategies including equipment operationalimprovements and upgrades, and occupant behavioral changes. These products often quantifywhole-building savings relative to a baseline period using methods that predict energyconsumption from key parameters such as ambient weather conditions and operation schedule.These automated baseline models streamline the M&V process and are of critical importance toowners and utility program stakeholders implementing multi-measure energy efficiencyprograms.This paper presents the results of a PG&E Emerging Technology program, undertaken toadvance capabilities in evaluating EMIS products for building-level baseline energy modeling. Ageneral methodology to evaluate baseline model performance was developed and used withhourly whole-building energy data from nearly 400 small and large commercial buildings.Evaluation metrics describing model accuracy were identified and assessed for theirappropriateness in describing model baseline performance, as well as their usefulness foridentifying and pre-screening buildings for whole-building savings estimation suitability. Thestate of five public-domain models was assessed using the methodology and test data set, andimplications for whole building M&V described. Finally a protocol was developed to test EMISvendor's proprietary models while navigating practical issues concerning test data security,vendor intellectual property, and maintaining appropriate testing blinds, while processing alarge data set. Ongoing work entails stakeholder vetting, demonstration of the test procedureswith new baseline models solicited from the public, and publication of the results for industryadoption.

Implementing energy conservation measures in buildings can reduce energy costs and environmental impacts, but such measures cost money to implement so intelligent investment strategies require the ability to quantify the energy savings by comparing actual energy used to how much energy would have been used in absence of the conservation measures (known as the baseline energy use). Methods exist for predicting baseline energy use, but a limitation of most statistical methods reported in the literature is inadequate quantification of the uncertainty in baseline energy use predictions. However, estimation of uncertainty is essential for weighing the risks of investing in retrofits. Most commercial buildings have, or soon will have, electricity meters capable of providing data at short time intervals. These data provide new opportunities to quantify uncertainty in baseline predictions, and to do so after shorter measurement durations than are traditionally used. In this paper, we show that uncertainty estimation provides greater measurement and verification (M&V) information and helps to overcome some of the difficulties with deciding how much data is needed to develop baseline models and to confirm energy savings. We also show that cross-validation is an effective method for computing uncertainty. In so doing, we extend a simple regression-based method of predicting energy use using short-interval meter data. We demonstrate the methods by predicting energy use in 17 real commercial buildings. We discuss the benefits of uncertainty estimates which can provide actionable decision making information for investing in energy conservation measures.