Oba Ramos, Caio Cesar [UNESP]de Sousa, Andra NunesPapa, João Paulo [UNESP]Falcao, Alexandre Xavier2014-05-202014-05-202011-02-01IEEE Transactions on Power Systems. Piscataway: IEEE-Inst Electrical Electronics Engineers Inc, v. 26, n. 1, p. 181-189, 2011.0885-8950http://hdl.handle.net/11449/8303Nowadays, 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.181-189engNontechnical lossesoptimum-path forestpattern recognitionArtigo10.1109/TPWRS.2010.2051823WOS:000286516100021Acesso restrito9039182932747194