Clustering Household Electricity Use Profiles

Abstract

An attempt was made to cluster the load profiles of a sample (n ≈ 380) of New Zealand households. An extensive range of approaches was evaluated, including the approach of clustering on "features" of the data rather than the raw data. A semi-automatic search of the problem space (cluster base, distance measure, cluster/partitioning method and k) resulted in a k = 3-cluster solution with acceptable quality indices and face validity. Although a particular combination of base, distance metric and clustering method was found to work well in this case, it is the practice of searching the problem space, rather than a particular solution, that is discussed and advocated.