Efficient iterative refinement clustering for electricity customer classification
Florentin Batrinu, Gianfranco Chicco, Roberto Napoli, Federico Piglione, Petru Postolache, M. Scutariu, Cornel Toader · 2005
Customer classification is aimed at providing to the electricity suppliers a sound information on the electricity consumption, to be used for formulating dedicated tariff structures. Different clustering methods can be adopted for assisting the process of electricity customer classification. Previous studies have identified two methods - hierarchical clustering and follow-the-leader - as most promising in terms of clustering validity for classifying the customers on the basis of the shape of their load patterns. However, the above methods exhibited some limitations in terms of the possibility of preassigning the number of clusters (for the follow-the-leader) or improving the cluster formation by reassigning the load patterns to the clusters already formed (for the hierarchical clustering). This paper presents the new iterative refinement clustering (IRC) method, originally developed in order to overcome these limitations. The performance of the IRC method has been compared to the one of other clustering methods by means of suitable clustering validity indicators. The results obtained on a set of over 200 non-residential customers are presented in the paper to show the effectiveness of the proposed IRC method.