We collected a building energy dataset (2 years, hourly data) of unprecedented size (3829 buildings) and variety (75 programs). And using machine learning, we discovered three fundamental load shape profiles that characterize the temporal energy use in any of the buildings. The existence of fundamental load shape profiles challenges the manmade, artificial classification of buildings. We demonstrate in a benchmarking application that the resulting data-driven groups are more homogeneous, and therefore more suitable for apple-to-apple comparisons of buildings.
This research is led by June in collaboration with Dr Miller’s BUDS Lab at NU Singapore.
Title: Apples or oranges? Identification of fundamental load shape profiles for benchmarking buildings using a large and diverse dataset
DOI: https://doi.org/10.1016/j.apenergy.2018.12.025
Free access until March 2nd: https://authors.elsevier.com/a/1YNnd15eiesNzd
The code for analysis is released on our github page: https://github.com/intelligent-environments-lab/ProfileClustering