Clasificación de fallas eléctricas aplicando redes neuronales artificiales a la protección de distancia de líneas de transmisión basada en el algoritmo de levenbergmarquardt

Abstract

This study presents a fault classification system based on artificial neural networks (ANN). In this sense, the types of faults considered for classification are phase-to-earth, phase-to-phase, three-phase and double line-to-earth faults. From another perspective, for ANN training, a data set is constructed, containing RMS values of voltages, fault currents and zero sequence currents, under different impedance and fault location parameters. These data are obtained from short-circuit studies and are used to extract the characteristics of the voltages and currents of each phase under normal and fault conditions. Therefore, the Levenberg-Marquardt algorithm is applied during the training phase of the ANN. For the validation of results, the fault classifier is tested using the IEEE 9 and 14 busbar test systems. From the tests performed, an average fault classification accuracy of 97% was obtained for each system.