Customer classification by means of harmonic representation of distinguishing features

This paper deals with grouping electricity customers on the basis of their electrical behavior. We refer to a set of daily load diagrams representing a macro-class of electricity customers defined on the basis of external classification criteria. We investigate on the selection of appropriate features to be used for grouping the customers by means of clustering techniques. We propose a novel characterization of the load diagrams by using a frequency-domain approach in which each customer is represented by a set of harmonics-based features. Assuming the day as fundamental period, we perform harmonic analysis of the representative load diagrams and define a set of shape factors by using magnitude and phase information of a relevant set of harmonics. We discuss some critical aspects of the feature selection, mainly linked to the treatment of the phase information. We show results of application of the proposed method to 232 customers of a real distribution system. Using the proposed harmonics-based features in a load diagram clustering procedure provides effective results, with a slight reduction in adequacy with respect to the use of 15-minute time-domain samples, well compensated by the significant reduction of the number of features used to represent the customer data and to run the clustering procedure.