Reduction and transformation of energy use data for end-user group categorization in dormitory buildings

• Published in: Journal of Building Engineering Vol. 32; p. 101524
• Main Authors: Song, Kwonsik, Ahn, Joseph, Ahn, Yonghan, Park, Moonsun, Kwon, Nahyun
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2020
• Subjects: Clustering analysis; Control strategy; Data preprocessing; Energy saving; Multi-zone buildings; Clustering analysis; Multi-zone buildings; Energy saving; Control strategy; Data preprocessing
• ISSN: 2352-7102; 2352-7102
• DOI: 10.1016/j.jobe.2020.101524

The control strategy of heating, ventilating, and air-conditioning (HVAC) systems is cost-effective to achieve energy saving in buildings. It is believed that identifying representative behavioral patterns of end-users in multi-zone buildings helps personalize the control parameters of HVAC systems. Thus, this will result in energy saving while minimizing thermal discomfort of end-users. With advanced metering technologies, it is possible to capture how end-users consume energy in their rooms and then categorize the rooms into several meaningful groups based on similar energy use patterns. Unfortunately, it is still unknown how changes in numerical values and dimension of energy use data affect the performance of end-user group categorization. Therefore, this research examines the performance of end-user group categorization across different types of numerical values and dimensions of energy use data. A clustering analysis is conducted using energy use data from 959 rooms of seven dormitory buildings in Seoul, South Korea. The clustering results show that reducing the dimension of energy use data (i.e., data reduction) improves the similarity of end-users within a group. Also, transforming the numerical values of energy use data (i.e., data transformation) makes the group similarity higher. Lastly, when combining both data reduction and transformation during the categorization process, the best clustering method is dependent on the distribution of energy use data. These results indicate that facility managers can provide end-users with thermally comfortable conditions and achieve energy saving across all zones. •We propose a process for end-user group categorization using unsupervised learning.•Various data reduction and transformation methods are used during the categorization.•Similarity among end-user group members increases with data reduction methods.•End-user groups have higher similarities by transforming energy use data.•The best clustering method varies depending on the distribution of energy use data.