This paper presents a general methodology to obtain a list of abnormal users using only the general customer databases as input. The abnormalities are oriented to fraud detection, customer consultancy purposes, and error correction (e.g. meter malfunction). This methodology allows the development of a system which can increase the success of the inspection campaigns by means of a systematic guidance to a reduced group of likely abnormal customers. This guidance is achieved by the characterisation of the consumption profiles of the customers by means of a non supervised Artificial Neural Network based algorithm. Patterns of «normal» consumption are extracted, each customer is assigned to one of them and is associated a normality degree which measures the similarity between the customer and the consumption patterns. Anomalous situations are detected either identifying users who do not fit any of the patterns, or identifying users who are associated with patterns which have been labelled as anomalous by the expert.
Keywords: Reduction of losses in distribution systems, abnormality and fraud detection, Artificial Neural Networks, data Mining
12th Conference of the Electric Power Supply Industry, Thailand, November 1998
Published: November 1998.