A Data-Driven Load Shape Profile Based Building Benchmarking: Comparing DOE Reference Buildings with a Large Metering Dataset

Abstract

There are various sustainable policies or labeling programs in building energy benchmarking. However, they are often oriented for the static information of building performances. The installation of smart metering capabilities with building energy system generates unprecedented amount of data. By analyzing such data, buildings can be evaluated in a more comprehensive manner. We propose a data analytic framework to discover dominant load shape patterns and benchmark buildings with respect to these. We applied this method to a simulation dataset (256 DOE reference building models) and compared them to the results of an actual metering dataset (3,829 buildings). Using k-means clustering, we found three fundamental profiles in the actual metering dataset, and two of them, i.e., noon and evening peak profile, are very similar to the dominant load shape patterns of the DOE reference buildings. After regrouping the buildings based on their dominant load shape patterns, we found that 94% of the buildings are assigned to one of the three fundamental profile shapes (in actual metering dataset), while there were only two groups assigned for the DOE reference buildings. The visual analytics of the proposed benchmarking provides a comprehensive insight on building performance, i.e., not only provides the static information (PSU and EUI) but also temporal aspects (load shape pattern) of building performance.