A New Approach for Nontechnical Losses Detection Based on Optimum-Path Forest

Abstract

Nowadays, fraud detection is important to avoid nontechnical energy losses. Various electric companies around the world have been faced with such losses, mainly from industrial and commercial consumers. This problem has traditionally been dealt with using artificial intelligence techniques, although their use can result in difficulties such as a high computational burden in the training phase and problems with parameter optimization. A recently-developed pattern recognition technique called optimum-path forest (OPF), however, has been shown to be superior to state-of-the-art artificial intelligence techniques. In this paper, we proposed to use OPF for nontechnical losses detection, as well as to apply its learning and pruning algorithms to this purpose. Comparisons against neural networks and other techniques demonstrated the robustness of the OPF with respect to commercial losses automatic identification.