DOE PAGES title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Comparison of One- and Two-Variable Linear Regression Models and Classic Energy Intensity for Energy Performance Tracking of Two Manufacturing Sectors

Journal Article · · Energy Engineering

We report manufacturing facilities consumed about 32% of total domestic energy in the United States in 2016. To evaluate the energy savings achieved through the implementation of energy conservation projects and to establish powerful arguments for future projects, accurate energy performance tracking methods are necessary. The classic energy intensity method (i.e., the ratio of annual total energy over annual total production) is the means most widely used to measure savings because it can be understood and calculated easily. This method considers the variation in production rates to some extent; however, it fundamentally assumes facilities' base energy consumption (energy consumption at zero production) to be zero, which rarely holds true. Furthermore, this method does not consider variations in other relevant parameters, such as weather conditions. Therefore, the regression models approach is commonly recommended to track energy performance improvements. However, because it requires more data and statistical expertise, the regression models approach has been adopted by only a few facilities, and many others suspect it is not worth the effort for their specific cases. For this reason, the improved accuracy this approach offers needs to be demonstrated. By analyzing 477 monthly energy (electricity and natural gas) data sets, this study quantitatively compared the accuracy of classic energy intensity, one-variable (production only) linear regression models, and two-variable (i.e., production and weather) linear regression models for two manufacturing sectors (primary metal and transportation equipment manufacturing). Results showed that significant improvements in accuracy were achieved with one-variable regression models when compared with the classic energy intensity method and with two-variable regression models when compared with one-variable regression models. These improvements were demonstrated using the p values of intercept, cooling degree days, and heating degree days. Based on these results, to achieve a good balance between accuracy improvements and resources requirements, one-variable (production only) linear regression models for electricity consumption and two-variable (production and heating degree days) linear regression models for natural gas consumption are recommended for both sectors in the case of limited resources. Lastly, facilities can also use the results to decide which approaches fit better in their cases.

Research Organization:
Oak Ridge National Laboratory (ORNL), Oak Ridge, TN (United States)
Sponsoring Organization:
USDOE Office of Energy Efficiency and Renewable Energy (EERE), Energy Efficiency Office. Advanced Manufacturing Office
Grant/Contract Number:
AC05-00OR22725
OSTI ID:
1474575
Journal Information:
Energy Engineering, Vol. 115, Issue 5; ISSN 0199-8595
Publisher:
Taylor & FrancisCopyright Statement
Country of Publication:
United States
Language:
English

References (7)

Applications of artificial neural-networks for energy systems journal September 2000
Measuring industrial energy savings journal May 2008
Measuring Progress with Normalized Energy Intensity journal April 2011
Artificial neural networks and physical modeling for determination of baseline consumption of CHP plants journal August 2014
A Method for Measuring the Efficiency Gap between Average and Best Practice Energy Use: The ENERGY STAR Industrial Energy Performance Indicator journal July 2005
Mathematical description for the measurement and verification of energy efficiency improvement journal November 2013
Industrial energy efficiency: the need for investment decision support from a manager perspective journal December 2003