Análise e detecção de faltas de alta impedância em sistemas de distribuição com a utilização de classificadores por aprendizado supervisionado

Abstract

The occurrence of high impedance faults (HIF) in primary distribution networks represents a danger to the safety of people, equipment, and animals. However, protection devices close to the distribution network are not able to be sensitized by this type of defect, in most cases. This work presents an integrated strategy for the classification and detection of HIF, based on the use of classifiers based on supervised learning algorithms, such as K-nearest neighbors, Support Vector Machines and Ensemble. An improved fault model was used to emulate random behaviors, and especially intermittency for high impedance faults. The residual current was monitored, and its characteristics were extracted using the Short-Time Fourier Transform (STFT). A temporal consistency logic was applied to the detection stage. The functioning of the presented algorithm was carried out through several simulations in a 20 kV distribution system.