Sampling frequency influence at fault locations using algorithms based on artificial neural networks

Abstract

A sampling frequency evaluation used in digital fault recorders for fault locations was implemented. A chained structure of artificial neural networks (ANN) was adopted to locate the faults. The ATP (Alternative Transient Program) software was used in the building of the database for training, testing and validation of the ANN, with different sampling frequencies. The input to the ANN are phase quantities and zero sequence voltage and current waveform data. The fault conditions were simulated for a 230 kV transmission line. The database used was generated automatically from a standard format file, and run in batch mode. For the fault location, the transmission line was divided into 8 zones. Previous to location, classification of the fault type is performed by training the ANN with the full line data. For the location, eight ANN were trained for each fault type, each one with the data of each zone.