Analysing energy use clusters of single-family houses using building and socio-economic characteristics

Clustering has been shown to be a promising approach to reduce the large amount of data from smart heat meters to representative profiles. However, attempts to understand why a case (building including its occupants) is within a particular cluster have only been moderately accurate. Therefore, this work uses existing energy use clusters based on about 4500 single-family homes to investigate whether socio-economic characteristics (SECs) alone or in combination with building characteristics (BCs) can improve the insight into the energy use clusters. An established variable selection and classification approach based on random forests was used. The results show that the eight SECs used alone provide poor insight into the energy use clusters, achieving only a Matthew Correlation Coefficient (MCC) of around 0.1. Simplifying the energy use clusters based on similarities, which was successful in the past, only moderately increased the MCC (≈ 0.17). When combined with BCs, SECs were never selected by the algorithm used, showing that they do not lead to a (significant) increase in MCC for both unsimplified and simplified clusters. Thus, this work suggests that SECs do not provide additional insights into why a case is within its respective energy use cluster.