Detecção e localização de faltas monofásicas em sistema de distribuição baseadas em redes neurais artificiais com uso de algoritmo genético para alocação de medidores

Abstract

In this work, a method utilizing artificial neural networks (ANNs) was proposed to detect and locate high-impedance faults in power distribution systems. Current measurements were used as input data for the ANNs, and to evaluate performance, three types of measurements were used: non-phasor, sequence, and phasor. Additionally, the genetic algorithm was employed to optimize the placement of the meters. The method was implemented by developing two sets of ANNs: one composed of neural networks associated with each bus of the distribution system, and another formed by three ANNs, each exclusive to a phase of the circuit, and this allowed for the detection of faults and the identification of defective buses and phases. The validation tests for the method were conducted on three systems: IEEE 13-bus, IEEE 34-bus, and IEEE 123-bus. The results showed that, despite the challenges imposed, the method was effective in most proposed scenarios. The ANNs proved to be robust in fault detection and defective phase localization, achieving efficiencies above 90% in most tested cases. The faulty bus location was indicated by a group of probable buses, which reduced the search field and showed efficiency above 90% when optimization was used. The placement of the meters was crucial for the method's performance, and the genetic algorithm proved to be an effective tool for this optimization. For the computational implementation of the method, the software OpenDSS® and Matlab® were used.