Identification of Electrical Measurement Irregularities and Prevention of Commercial Losses in Billing

Abstract

This work presents the development of a methodology capable of identifying discrepancies in the billing, measurement, and real electricity consumption of medium voltage customers of the energy distribution company in Rio de Janeiro, using Machine Learning techniques based on the analysis of consumption curves, demand, and load factors of consumer units (CU) and other exogenous information present in the database. The methodology is based on two stages: categorization or clustering, to group together consumer units with similar consumption patterns; and classification, to discover changes in the customer's behavior profile, configuring irregularities in electricity metering. The grouping of records is one of the tasks carried out in the Data Mining process in such a way as to reflect the structure of a data set in groups (clusters). A methodology was therefore developed based on the Fuzzy C-Means (FCM) algorithm for grouping data in cases where the number of clusters is known a priori. For classification, a new method is proposed that allows capturing the natural variation in energy consumption, as well as the effects of seasonal patterns on consumption, enabling groups to move based on time and energy consumption to avoid false detection of irregularities. The data from the customer base used for testing was initially pre-processed and normalized, with the intention of increasing the accuracy of the method. The results obtained with modeling using computational intelligence techniques made it possible to identify new forms of irregularities in consumption and increase energy recovery, validating the results with field inspections.