Selection of Suitable Features for Fault Classification in Transmission Line

Power system protective relaying is an important feature for efficient and reliable power flow operation. The traditional fault classification scheme using the steady state component is easily affected by environment factors. Then, it is important to provide power protection scheme that offers a better classification performance based on fault-generated transient signals where it is immune to the surrounding factors. This paper search for important features for fault classification in transmission lines using Wavelet Transform (WT) and multilayered perceptron (MLP) network. Six (6) features namely wavelet energy, mean, standard deviation, entropy, kurtosis and skewness are obtained from the WT. For analyzing these features, a MLP network trained by Levenberg-Marquardt (LM) algorithm is used as classifier to identify the fault types. The classification accuracy is evaluated using three types of dataset conditions; ideal dataset (no noise involvement), dataset with Signal-to-noise ratio (SNR) of 30 (30 dB noise) and dataset with SNR of 20 (20 dB noise). Simulation results show that combination of the wavelet mean and standard deviation shows the highest performance accuracy for all conditions tested.