Detection and Tracking of Atypical Billings Aiming at Minimizing Non-Technical Losses in Distribution Companies in Brazil

The work addresses the development of a computational platform to identify atypical billings in low voltage consumer units using Artificial Intelligence techniques, specifically the Isolation Forest algorithm. The proposed methodology allows detecting consumer units with peculiar changes in consumption and/or evidence of fraud. The algorithm analyzes a set of historical data and identifies “atypical” monthly billings among consumer units with the same load profile, generating a list of candidate units for subsequent inspection in the field. Preliminary tests carried out in a medium-sized distribution company showed an assertiveness rate of 91.70%, indicating the effectiveness of the proposed algorithm for identifying anomalies in the monthly billing data of consumer units.