Application of extreme learning machine for underground cable fault location

Abstract

This paper presents an accurate hybrid fault location technique, combining s-transform and extreme learning machine for underground power cable in distribution system. In the proposed method, one cycle sending end post fault current and voltage signal are taken for determining the fault location in an underground cable. Using s-transform, useful features are extracted, and further applying feature selection technique, redundant features are removed from the total feature set. In this paper, forward feature selection/particle swarm optimization-based feature selection method is used. Thereafter extreme learning machine is used to estimate the fault distance with the selected features. Feasibility of the proposed method has been tested for all ten types of fault on a 20-kV, 5-km underground power cable and 220-kV, 10-km underground cable with a large range of operating condition. Also the proposed method is evaluated for 220-kV, 10-km underground cable in combination with 100-km overhead line. The simulation result of the proposed fault location technique shows that the maximum absolute error of less than 0.3% and a mean error of less than 0.2% are achieved which demonstrate high accuracy and robustness. Results are compared with other fault location approaches. Copyright © 2014 John Wiley & Sons, Ltd.