Exploring smart heat meter data: A co-clustering driven approach to analyse the energy use of single-family houses

The ongoing digitalisation of the district heating (DH) sector opens new doors for data-driven methods. Remotely readable meters (smart meters) create data at an unprecedented extent and temporal resolution, which allows gaining inside into the energy use of buildings. This insight can support the needed renovation wave and the transformation of the current DH networks to low-temperature 4th generation DH networks. This work contributes by proposing a novel workflow to establish energy use clusters without relying on fixed season definition, addressing the challenge that climate change makes static seasonal definitions difficult to establish. Further, these clusters are analysed regarding their relationship with respect to 26 building characteristics (BCs) to understand why a building is within a specific cluster using classification and variable selection methods. The results, based on two years of data of 4798 single-family houses, show that the used co-clustering approach establishes well-separated energy use clusters. While correlated to the exterior temperature, the found season variation does not follow commonly used fixed season definitions and further varies across years, showing that fixed season definitions do not correctly capture the individual seasonal variation of energy use in single-family houses. The results of the variable selection and classification approaches show that even highly detailed BCs are not sufficient to explain why a building is in its respective cluster (Matthew’s correlation coefficient (MCC) ≈0.3). By artificially simplifying the found energy use clusters based on similarities in their energy use profile and magnitude, the performance of the classification could be significantly increased (MCC ≈0.5). For both the not simplified and simplified energy use clusters, simple BCs, which are inexpensive to collect and in most cases, already available, lead to a similar understanding as detailed BCs.