A regressionbased approach to estimating retrofit savings using the Building Performance Database
Abstract
Retrofitting building systems is known to provide costeffective energy savings. This article addresses how the Building Performance Database is used to help identify potential savings. Currently, prioritizing retrofits and computing their expected energy savings and cost/benefits can be a complicated, costly, and an uncertain effort. Prioritizing retrofits for a portfolio of buildings can be even more difficult if the owner must determine different investment strategies for each of the buildings. Meanwhile, we are seeing greater availability of data on building energy use, characteristics, and equipment. These data provide opportunities for the development of algorithms that link building characteristics and retrofits empirically. In this paper we explore the potential of using such data for predicting the expected energy savings from equipment retrofits for a large number of buildings. We show that building data with statistical algorithms can provide savings estimates when detailed energy audits and physicsbased simulations are not cost or timefeasible. We develop a multivariate linear regression model with numerical predictors (e.g., operating hours, occupant density) and categorical indicator variables (e.g., climate zone, heating system type) to predict energy use intensity. The model quantifies the contribution of building characteristics and systems to energy use, and we use it to infermore »
 Authors:
 Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States). Energy Analysis and Environmental Impacts Division; Univ. of California, Berkeley, CA (United States). Civil and Environmental Engineering Dept.
 Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States). Energy Analysis and Environmental Impacts Division
 Publication Date:
 Research Org.:
 Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)
 Sponsoring Org.:
 USDOE Office of Science (SC)
 OSTI Identifier:
 1378362
 DOE Contract Number:
 AC0205CH11231
 Resource Type:
 Journal Article
 Resource Relation:
 Journal Name: Applied Energy; Journal Volume: 179; Journal Issue: C
 Country of Publication:
 United States
 Language:
 English
 Subject:
 32 ENERGY CONSERVATION, CONSUMPTION, AND UTILIZATION
Citation Formats
Walter, Travis, and Sohn, Michael D. A regressionbased approach to estimating retrofit savings using the Building Performance Database. United States: N. p., 2016.
Web. doi:10.1016/j.apenergy.2016.07.087.
Walter, Travis, & Sohn, Michael D. A regressionbased approach to estimating retrofit savings using the Building Performance Database. United States. doi:10.1016/j.apenergy.2016.07.087.
Walter, Travis, and Sohn, Michael D. 2016.
"A regressionbased approach to estimating retrofit savings using the Building Performance Database". United States.
doi:10.1016/j.apenergy.2016.07.087.
@article{osti_1378362,
title = {A regressionbased approach to estimating retrofit savings using the Building Performance Database},
author = {Walter, Travis and Sohn, Michael D.},
abstractNote = {Retrofitting building systems is known to provide costeffective energy savings. This article addresses how the Building Performance Database is used to help identify potential savings. Currently, prioritizing retrofits and computing their expected energy savings and cost/benefits can be a complicated, costly, and an uncertain effort. Prioritizing retrofits for a portfolio of buildings can be even more difficult if the owner must determine different investment strategies for each of the buildings. Meanwhile, we are seeing greater availability of data on building energy use, characteristics, and equipment. These data provide opportunities for the development of algorithms that link building characteristics and retrofits empirically. In this paper we explore the potential of using such data for predicting the expected energy savings from equipment retrofits for a large number of buildings. We show that building data with statistical algorithms can provide savings estimates when detailed energy audits and physicsbased simulations are not cost or timefeasible. We develop a multivariate linear regression model with numerical predictors (e.g., operating hours, occupant density) and categorical indicator variables (e.g., climate zone, heating system type) to predict energy use intensity. The model quantifies the contribution of building characteristics and systems to energy use, and we use it to infer the expected savings when modifying particular equipment. We verify the model using residual analysis and crossvalidation. We demonstrate the retrofit analysis by providing a probabilistic estimate of energy savings for several hypothetical building retrofits. We discuss the ways understanding the risk associated with retrofit investments can inform decision making. The contributions of this work are the development of a statistical model for estimating energy savings, its application to a large empirical building dataset, and a discussion of its use in informing building retrofit decisions.},
doi = {10.1016/j.apenergy.2016.07.087},
journal = {Applied Energy},
number = C,
volume = 179,
place = {United States},
year = 2016,
month = 7
}

Ambienttemperature regression analysis for estimating retrofit savings in commercial buildings
This paper describes a procedure for estimating weatheradjusted retrofit savings in commercial buildings using ambienttemperature regression models. The selection of ambient temperature as the sole independent regression variable is discussed. An approximate method for determining the uncertainty of savings and a method for identifying the data time scale which minimizes the uncertainty of savings ar developed. The appropriate users of both linear and changepoint models for estimating savings based on expected heating and cooling relationships for common HVAC systems are described. A case study example illustrates the procedure. 
Uncertainty in baseline regression modeling and in determination of retrofit savings
The objective of this paper is to discuss the various sources of uncertainty inherent in the estimation of actual measured energy savings from baseline regression models, and to present pertinent statistical concepts and formulae to determine this uncertainty. Regression models of energy use in commercial buildings are not of the standard type addressed in textbooks because of the changepoint behavior of the models and the effect of patterned and nonconstant variance residuals (largely as a result of changes in operating modes of the building and the HVAC system). This paper also addresses such issues as how model prediction is impactedmore »