Um localizador de faltas de alta impedância a partir de dados reais utilizando inteligência artificial.

Abstract

Due to the radial power distribution systems topology, a large number of consumers may remain without power supply after the operation of the protection system. Thus, an estimation of the fault occurrence place represents an important step in reducing outage time for consumers. There is a disturbance type known as High Impedance Faults (HIF), which usually occur when there is a drop or energized conductor contact of the primary circuit network with a high value resistive surface, like trees, roads or buildings. In this case, the protection system is not sensitized. The HIF have specific characteristics, such as asymmetry and nonlinearity. The goal here is to develop an algorithm to perform the HIF location in distribution systems using Artificial Neural Networks (ANN) and Wavelet Transform. The proposed location is performed in two steps, using two ANN classes: MLP - Multilayer Perceptron, for extension of the location and SOM - Self Organizing Maps, to determine the sector feeder. The used input data are the phase currents, as well as the energy of the wavelet coefficients. For HIF modeling, field tests were carried out. The results indicated an accuracy of around 90%, indicating the possibility of using Artificial Intelligence to the problem.