Classificação de eventos em redes de distribuição de energia elétrica utilizando modelos neurais autônomos

Abstract

This work presents a method for automatic classification of faults and events related to quality of service in power distribution networks, based on oscillographies of the bar feeder voltages of the distribution substation. We present the results for two distinct pre-processing forms of the voltage signals. The first is based on the Fourier Transform and the second on the Wavelet Transform for different families of wavelet functions. We compared three neural models for the process of classification: Multi-Layer Perceptron, Radial Basis Function and Support Vector Machine. The models were trained taking into account the autonomous operation of networks, i.e. automatic model selection and control complexity. The results were validated for a set of simulations performed using the Alternative Transient Program, aimed at practical implementation of the proposed method in an oscillograph logger, developed by Lactec together with Copel, called the Power Quality Monitor. The results were obtained with performance on the order of 90% of average accuracy for the various pre-processing forms and neural models.