Search eceee proceedings

Using machine learning and mathematical programming to benchmark energy efficiency of buildings

Panel: 5. Buildings and construction technologies and systems

This is a peer-reviewed paper.

Authors:
Alexander Severinsen, Storekeeper AS, Norway
Helen Sorensen Holst, School of Business and Economics, Norway

Abstract

In this paper, we demonstrate a novel benchmarking technique to identify buildings with a potential to reduce energy consumption, taking into account both weather effects and building characteristics. The proposed method will quantify if a building's energy usage is dependent on outside temperature, and the degree of dependency over time. Our method prepares the data for analysis in a way that can improve the results from benchmarking techniques. A demonstration is performed with the use of two different data envelopment analysis (DEA) models. The first DEA model without taking into account temperature, and a second where temperature is included as a non-controllable variable. Results suggest that the average store is 28% less efficient than the most efficient stores. Further, our analysis show that using our suggested analytical framework will improve the accuracy of the efficiency scores compared to more standard methods. This is an important finding suggesting that our proposed methodology has advantages over existing benchmarking methods.

Downloads

Download this paper as pdf: 5-157-17_Severinsen.pdf

Download this presentation as pdf: 5-157-17_Severinsen_display.pdf