A holistic time series-based energy benchmarking framework for applications in large stocks of buildings

Abstract

With the proliferation of Internet of Things (IoT) sensors and metering infrastructures in buildings, external energy benchmarking, driven by time series analytics, has assumed a pivotal role in supporting different stakeholders (e.g., policymakers, grid operators, and energy managers) who seek rapid and automated insights into building energy performance over time. This study presents a holistic and generalizable methodology to conduct external benchmarking analysis on electrical energy consumption time series of public and commercial buildings. Differently from conventional approaches that merely identify peer buildings based on their Primary Space Usage (PSU) category, this methodology takes into account distinctive features of building electrical energy consumption time series including thermal sensitivity, shape, magnitude, and introduces KPIs encompassing aspects related to the electrical load volatility, the rate of anomalous patterns, and the building operational schedule. Each KPI value is then associated with a performance score to rank the energy performance of a building according to its peers. The proposed methodology is tested using the open dataset Building Data Genome Project 2 (BDGP2) and in particular 622 buildings belonging to Office and Education category. The results highlight that, considering the performance scores built upon the set of proposed KPIs, this innovative approach significantly enhances the accuracy of the benchmarking process when it is compared with a conventional approach only based on the comparison with the buildings belonging to the same PSU. As a matter of fact, an average variation of about 14% for the calculated performance scores is observed for a testing set of buildings.