Clustering-Based Assessment of Residential Consumers from Hourly-Metered Data

This paper addresses the methodology for determining suitable groups of residential consumers, based on time series of their hourly energy consumption and contractual data. Salient aspects are the discussion on the importance of the data representation in terms of data normalisation, choice of the appropriate features to be used as inputs in clustering procedures, and computation of clustering validity indicators. The analysis is carried out on real hourly-metered electricity consumption data of 10,000 residential consumers. We discuss the main insights obtained with the application of conventional approaches based on time series data handled with different distance metrics (e.g., Euclidean distance and dynamic time warping) and alternative approaches exploring data transformations, among which the CONsumption DUration Curve Time Series (CONDUCTS) methodology proposed by the authors.