Artificial neural network and Stockwell transform for fault location in transmission lines

• Main Author: Saulo Cunha AraÃjo de Souza
• Publication Date: 2015
• Format: Master thesis
• Language: por
• Source: Biblioteca Digital de Teses e Dissertações da UFC
• Download full: http://www.teses.ufc.br/tde_busca/arquivo.php?codArquivo=15403
• topic: Linhas de transmissÃo; Redes neurais artificiais; Transformada de Stockwell; Teoria das ondas viajantes; LocalizaÃÃo de faltas; Transmission lines; Artificial neural network; Stockwell transform; Travelling waves theory; Fault location; SISTEMAS ELETRICOS DE POTENCIA

This paper presents an automatic fault location method in transmission lines based on the Travelling Waves Theory (TWT) using the Stockwell Transform (ST) to determine the travelling waves propagation time and the dominant frequency of transient signals generated by faults. The method considers the case where there is no communication between terminals or loss of synchronism between the devices responsible for estimating the location of faults using, therefore, only data from one terminal. Single-phase faults only involving one of the phases and the earth area evaluated, which occur in the first half of a transmission line of unknown parameters. It is observed that the method (i) wasnât sensitive to fault resistance variations and inception angle and (ii) the obtained results presented errors between 0,10% and 5,82% for faults that occurred between 7km and 99km from the monitoring terminal. To improve the accuracy of estimating the fault location, an Artificial Neural Network (ANN) of the type MLP (Multi-Layer Perceptron) is designed, and trained with characteristics extracted from the faulty signals using ST. The ATP (Alternative Transient Program) software was adopted for simulation of a three phase transmission line which voltage signals were sampled at 200kHz. The simulations were performed exploring 1280 combinations of the following parameters: fault locations, fault resistances and inception angle. The method was developed using the software MATLABÂ. According to the obtained results, the combination of ST with ANN presented better results than the application of ST and TWT. Such improvement is highlighted for the estimation of fault location at greater distances from the monitoring terminal, with errors between 0,02% and 1,56% for faults that occurred between 7km and 99km from the monitoring terminal.