Cluster analysis of energy consumption mix in the Japanese residential sector

Abstract

Sector integration is one of the major components considered when designing future smart energy systems. Due to the lack of data, current partitioning of consumers into residential, commercial, industrial, and transportation sectors is based on assumptions on their respective energy demand. In reality though, there is a diversity in preferences and behaviors so that consumers from the same sector may have different demand and consumers from different sectors may have similar demand. With the increasing amount of individual data, it is becoming possible to study this diversity and more accurately partition consumers using advanced analysis techniques such as clustering. However, while this approach does allow for an accurate grouping, the complex mechanisms at the roots of identified clusters are still unclear. Indeed, energy demand depends on multiple factors such as economic, political, cultural, social, or historical besides environmental conditions. The present study uses households' data provided by the Japanese Ministry of the Environment with the objective of finding patterns associated with residential energy consumption profiles. It is found that the probability distribution of households' energy consumption seems to be log-normal so that clusters are revealed by first applying a logarithmic nonlinear transformation. Furthermore, k-means clustering, which is commonly used in energy systems study, fails here to correctly identify the clusters when compared with density-based clustering. After identifying clusters, we look for statistically significant specificities in the corresponding households' data such as their geographical location, number of residents, income, buildings' construction year, equipment and vehicles and suggest interpretations for each.