Fuzzy detection of high impedance faults in radial distribution feeders

Abstract The paper presents a methodology for the detection of high impedance faults in radial distribution feeders. The technique consists of making comparative analysis of the responses of the feeder to pulses injected at the feeder inlet for different configurations. The responses to normal configurations constitute the set, hereafter called ‘standard responses’, of the feeder under test. An artificial neuron set, composed of ‘neo-fuzzy’ neurons, is trained to recognize the standard responses. After the training period, the neuron-set becomes the decision core of the supervisory system. Subsequently, the supervisory system is used to monitor the feeder in real time. The feeder responses are periodically analysed by the neuron-set, which classifies them according to the ‘pre-defined’ standard responses. In this way, non-standard configurations are not classified. Simulation and experimental studies have shown that the proposed system is capable of recognizing passive high impedance faults even when they occur close to a switch (used to change configuration). In order to corroborate the efficiency of the proposed methodology, experimental tests have been accomplished in a real feeder. They have shown satisfactory results.