Localização de faltas no sistema elétrico de distribuição da Universidade Federal da Grande Dourados utilizando inteligência artificial

Abstract

This work proposed a methodology for locating faults through Artificial Neural Networks (RNAs) for the distribution power system electrical of the Federal University of Grande Dourados (UFGD). A brief theoretical survey was carried out on existing fault localization techniques and on the main properties of RNAs. For the computational modeling of the electrical system, the ATPDraw software is used, showing itself a complete tool, allowing simulation with too many distribution lines (LTs), transformers and loads of the system, where a total of 48 threephase short-circuits were simulated. For the treatment of data, for the purpose of forming the RNA input matrix, to make a base code for extracting the root mean square (RMS) of the current and voltage data of a certain time interval. Five input arrays were assembled to test one location effectiveness with one or more terminals in the modeled network. Using the Neural Network Pattern Recognition Tool (nprtool) of the MATLAB software, it was found that a better form of fault localization with an applied methodology is the data reference of three measurement terminals.