Comparison of clustering approaches for domestic electricity load profile characterisation - Implications for demand side management. (1st August 2019)

• Record Type: Journal Article
• Authors: Yilmaz, S.; Chambers, J.; Patel, M.K.
• Is Part Of: Energy. Volume 180(2019)
• Page Start: 665
• Page End: 677
• Publication Date: 2019-08-01
• Subjects: Clustering -- K-means -- Electricity load profiles -- Features -- Smart-meters -- Demand side management; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2019.05.124
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Cluster analysis is increasingly applied to smart meter electricity demand data to identify patterns in electricity consumption in order to improve load forecasting and to enhance targeting of demand response programmes. The analysis was performed on one year of smart meter electricity demand data from 656 households in Switzerland. We present a rigorous assessment of sample aggregation and clustering approaches for creating representative electricity demand profiles. We propose a clustering method using five features defining the shape of household electricity demand profiles, which demonstrates significantly improved cluster quality relative to using raw profile data. The cluster analysis of average household electricity demand profiles resulted in three distinct clusters, which challenges the assumption made by Swiss energy norms that one standard pattern fits all homes. Furthermore, cluster analysis of daily demand profiles within the household was performed, resulting in four distinct clusters and demonstrating that daily raw profiles for a household significantly differ from the average profile for that household. Averaging the data suppresses the diversity of the electricity use patterns within the individual household. Electricity demand profiles have important implications for policy makers, particularly if time of use tariffs are introduced to match future stochastic renewable energy supply. Highlights: Provides a rigorous assessment on clustering and sample aggregation approaches. Feature based clustering approach improves the quality of the clusters. Averaging the data suppresses the diversity of the electricity use patterns. Provides a tailored analysis for selected applications of demand side management. Theoretical and practical implications for demand response tariffs are provided.