Transmission Line Fault Location Using MFCC and LS-SVR

The location of Transmission Line (TL) Faults is a major problem in Electrical Power Systems (EPSs), since precisely identifying the point of occurrence of a fault in a TL it is possible to perform a faster restoration of the operation to the desired normal conditions. In this work we used a Least-Squares Support Vector Regression (LS-SVR) to locate faults in a TL with inputs provided by MFCC (Mel-Frequency Cepstral Coefficients) obtained from voltage signals during the fault. A modelled line based on parameters of a real line was used, with a total of 4008 fault situations being simulated on this Transmission Line. It is important to point out that MFCC are not used in applications involving EPS’s, and, according to the bibliographic research conducted by the team so far, no application of this feature extraction tool has been detected for the TL fault location problem. 3006 faults were used to train the model with cross-validation by the k-fold method, and 1002 faults were used for testing. The proposed methodology presented a good performance in the tests carried out, with a mean relative error of 0.000419 ±0.000640% when models are trained and tested with noiseless voltage signals. For models trained with voltage signals that present SNR ranging from 100 dB to 25 dB, the relative mean error ranged from 0.00334 ±0.00459%, in the first case, to 0.030580±0.043160% in the last.