Clustering Household Electrical Load Profiles Using Elastic Shape Analysis

Quantification and detection of patterns in electricity consumption curves, across households, locations, and seasons, is crucial for planning and forecasting. Treating daily consumption load curves as functional data, this paper utilizes techniques from elastic shape analysis to cluster and analyze load curves according to their shapes. The key idea is to time warp load curves in order to align their peaks and valleys using an elastic shape metric. This step removes the phase variability from the data and helps to focus on the shapes of load profiles. The resulting shape metric is then used to find dominant clusters in the household data using a Bayesian nonparametric clustering algorithm. The cluster averages correspond to the predominant usage patterns at the household level and help discover broad patterns across neighborhoods and seasons. This framework is demonstrated using actual electricity consumption data from the city of Tallahassee, FL, USA.