Comparing time-series clustering approaches for individual electrical load patterns

Abstract

This work positions the task of grouping electricity load time series among the vast field of clustering, and highlights corresponding research issues. A selection of the most performant time-series clustering approaches from the signal processing community are compared on the same dataset, composed by domestic electricity load profiles from Spain. The cross-correlation-based distance of Paparrizos and Gravano (2015) is shown to provide the best tradeoff between clustering accuracy and CPU times. © 2017 The Institution of Engineering and Technology. All rights reserved.