Proteção diferencial em transformadores de potência via redes neurais artificiais

Abstract

The present research is a study about the differential protection in power transformers, using neural networks. This research has as general objective, to identify internal and external faults of a power transformer, in several operating conditions, using artificial neural networks (RNA). The ATPDraw software was used to model and analyze the behavior of a 25MVA power transformer in a variety of operating conditions such as: permanent regime, internal phase-to-earth faults in the primary of the transformer and near-load faults. Among the highlighted conditions, the following stand out: situations of internal faults, external faults, permanent regime. An RNA network was implemented in Matlab, in order to classify the conditions of simulated operations. The implemented RNA architecture was a multi-layer FeedForward Network, the Multilayer Perceptron. The research found that using the simulation data generated in ATPDraw as input data for RNA, satisfactory results were obtained, in which the algorithm correctly classified a considerable percentage of samples, showing a possible alternative method of differential protection to conventional algorithms.