Assistente baseado em redes neurais artificiais para indicação de clientes suspeitos de fraude em sistema de distribuição de energia elétrica de baixa tensão.

Abstract

The search for measurement points with irregularities in energy distribution systems has shown itself to be a costly task when accomplished through non-automated processes, which require great amounts of time and physico-mental effort of those involved. These costs are increased for low-voltage services, where the collection of energy consumption readings are field-based and there is no monitoring of several other electrical quantities through telemetry. This work, then, implements an assistant based on artificial neural networks theory for detecting irregularities (potential frauds) of monthly energy measurements. With basis in historical consumption, fraudulent and non-fraudulent clients are classified with a success rate of up to 83.80%, utilizing a multilayer perceptron artificial neural network trained with the resilient backpropagation algorithm. The outcomes presented in this work can be further utilized to approach the fighting of nontechnical losses caused by fraudulent means, with neural networks in electrical energy distribution systems.