A comparative Study Between Electrical and Morphological Features for Short-Circuit Faults Detection and Discrimination in Power Grid Lines

Abstract

This paper outlines the adopted methodology to construct an intelligent system, which is able to detect and to discriminate between short circuit faults in high voltage power lines (220 kV, 50 Hz) with a length of 300 km. Based on the current study, two approaches for feature extraction are presented and compared. Firstly, the voltage and current signals are decomposed into 20 ms segments, and two distinct sets of descriptors are then calculated; The first one, consists on a set of 102 morphological, and the second one, consists on a set of 12 electrical parameters. Finally, two direct probabilistic multiclass support vector machines (M-SVM) are trained separately to discriminate between 10 short-circuit faults plus a normal case, each of them receives as inputs one of the previously calculated sets.The study shows that the obtained results are very satisfactory, however, the M-SVM presents higher accuracy when it’s trained by morphological parameters; with a classification rates of 96.74% and 91.23% for the first and second method respectively