Using neural network techniques for identification of high-impedance faults in distribution systems

The objective of the present paper is to present a multi-parametric approach based on artificial neural networks for identification and classification purposes of high-impedance faults in distribution systems. More specifically, the proposed methodology uses artificial neural networks integrated with other several statistical techniques that have also been used in these problem types. Besides providing a modular and robust architecture under the point of view of fault occurrences, the developed intelligent system was implemented using efficient tools dedicated to the preprocessing procedures of voltage and current signals registered from the substation. The global efficiency of the preprocessing tools is guaranteed because part of them accomplishes inferences in the time domain, while the other infer results using the frequency domain. The final results obtained from the application of the proposed system were fully successful, having the same ones tested and validated in distribution feeders from voltage and current signals registered in the substation, which are involved with real fault situation.