The effect of linear regression modeling approaches on determining facility wide energy savings
Holistic energy management system business practices, such as the framework detailed in ISO 50001 – Energy management system, requirements, and guidance for use, are centrally based upon the concept of energy performance improvement. For the purposes of ISO 50001, energy performance improvement can be determined for boundaries and following a process that is best suited for the implementing organization. Many organization and government ISO 50001 based programs, including the United States Department of Energy Superior Energy Performance (SEP) program, find value in demonstrating organizational energy performance improvement as the difference in energy consumption between two time periods within physically defined facility-boundaries.
To make the difference in energy consumption for the two time periods meaningful, the amount of energy consumed must be adjusted to account for relevant variables. Relevant variables, such as metrics of production and weather conditions, affect directly the amount of energy consumed but are independent of the facility’s energy performance. Therefore, it is crucial to adjust the observed energy consumption by the relevant variables identified by the facility, so that the energy savings resulted from the energy performance improvement actions can be isolated and determined, which is the purpose of this program.
To tackle this adjustment issue, the SEP measurement and verification (M&V) protocol specifies four energy consumption adjustment modeling methods for use; forecast, backcast, standard conditions, and chaining. Application of a single set of energy consumption and relevant variable data from a manufacturing facility to the four different energy consumption adjustment modeling methods produces four different energy savings values. Variation in the energy savings values is the result of inevitable changes in operation and conditions between the baseline and reporting periods, which affects the evaluation results significantly. The lack of agreement in the calculated energy savings values, while all meeting the requirements of the SEP M&V Protocol, indicates that additional context and analysis is required to understand which modeling method, and subsequent result, best represents the actual energy performance improvement of an organization.
This report describes how each adjustment model method can be implemented and provides guidelines of how to choose an appropriate adjustment method. A variety of statistical tests were made use to reveal which of the four methods best reflects the energy performance improvement of a given organization. In the study case of this paper, all of the four adjustment methods were applied. The resulting four savings estimates ranging from -1091.4 to 142,248.0 MMBtu, and the four SEP Energy Performance Indicator (SEnPI) estimates ranging from 0.93 to 1.00 lead to drastically different energy performance improvement conclusions. Discussion focused on why an organization should pick one of the four modeling methods over the others is carried out. An average energy saving estimator was proposed in order to provide a more appropriate estimate for the ESP. The interpretation of the 95% confidence intervals associated to ESP may provide more context to what calculated energy savings values mean and how they should be interpreted.