An Energy Performance Benchmarking of office buildings: A Data Mining Approach

Abstract

The COVID–19 pandemic has affected the world economy and is likely to have a dramatic impact on the world’s clean and sustainable energy. Focused efforts to improve the energy efficiency of buildings have been and will be even more essential to achieve desired sustainability goals. The energy benchmarking enables an understanding of the relative energy efficiency of buildings and identifying potential energy saving opportunities. In this sense, this paper aims to develop an energy performance benchmark for office buildings using data mining techniques that have been widely used in literature, showing robustness and reliability results. Specifically, we used techniques such as a wrapper model based on regression analysis for feature selection and the K-prototypes algorithm for classifying buildings. The key idea is to cluster the buildings containing mixed-type data (both numeric and categorical) and establish a benchmarking in each group according to the relative significance (weight) of each building. As a result, eight types of energy benchmarks were developed for each cluster of office buildings, and these were validated in terms of Adjusted R-squared. The results showed that the proposed approach outperformed the Energy Star method by 18%.