Detection and Identification of Abnormalities in Customer Consumptions in Power Distribution Systems

Abstract

This paper proposes a computational technique for the classification of electricity consumption profiles. The methodology is comprised of two steps. In the first one, a C-means-based fuzzy clustering is performed in order to find consumers with similar consumption profiles. Afterwards, a fuzzy classification is performed using a fuzzy membership matrix and the Euclidean distance to the cluster centers. Then, the distance measures are normalized and ordered, yielding a unitary index score, where the potential fraudsters or users with irregular patterns of consumption have the highest scores. The approach was tested and validated on a real database, showing good performance in tasks of fraud and measurement defect detection.