This paper explores a new approach for clustering time-series electricity load data through the development of map models. The approach provides an avenue for clustering load profiles that are noisy, have gaps or are non-overlapping. Performance is compared against an alternative strategy for clustering based on the extraction of time-series features through principal component analysis.

Results examine the clustering of thousands of residential load curves made publicly available through the Smart Grid Smart City program conducted by Ausgrid (see https://bit.ly/2toz2r0).

Note that this clustering approach has also been deployed with NEAR Program data assets (see "Victorian residential load behaviours" in Related Assets below).