Computational Intelligence Methods Applied to the Fraud Detection of Electric Energy Consumers

Abstract

The electric energy loss due to irregularities, fraud and
robbery represents a loss of revenue of billions of reais (Brazilian
currency) for Brazilian distributors every year. Several approaches
have been used to combat non-technical losses by energy utilities,
such as campaigns for prevention, raising awareness of the
population and encouraging the denunciation of fraudsters. A
strategy that is gaining strength is conducting technical inspections
at consumers to whom there is some kind of evidence, based on
customer data, consumption patterns, reader's notes, etc. This
paper proposes the application of neural network models on
selecting a set of consuming units for which there is a high
probability of frauds and irregularities. First, concepts related to
non-technical losses of energy are described and the types of
consumer data to which the companies usually have access. The
network input features are selected and some training models are
constructed. The resulting models are compared considering the
precision, accuracy and duration of training.