-
Felipe Munaro Lima
Programa de Pós-Graduação em Engenharia Elétrica e Telecomunicações, Universidade Federal Fluminense, Niterói – RJ
-
Andre Abel Augusto
Departamento de Engenharia Elétrica, Universidade Federal Fluminense, Niterói – RJ
-
Vitor Hugo Ferreira
Departamento de Engenharia Elétrica, Universidade Federal Fluminense, Niterói – RJ
Keywords:
Fault Diagnosis, Phase Shift Power Transformers, Machine Learning, Information Theory, Artificial Neural Network
Abstract
Machine Learning techniques for fault location and diagnosis in phase shift power transformers have been proposed in the literature. However, most of them adopt loss functions that consider a few moments of the error distribution, not adequately extracting the information stored in data available for diagnosis. Information theoretic criteria can overcome this limitation, resulting in better training and inference, and consequently, enhancing the fault analysis. This work presents an information theoretic learning approach to fault analysis in indirect symmetrical phase-shift transformers. The methodology consists of training an artificial neural network for fault detection using information theoretic cost functions. A comparative study with the traditional mean square error cost function will also be present.