RBF Neural Network-Based Wavelet Packet Energy-Aided Fault Localization on a Hybrid Transmission Line

Abstract

This paper presents a fault localization technique based on wavelet packet decomposition (WPD) and radial basis function neural network (RBFNN) for a hybrid transmission line consisting of an overhead line and an underground cable fed from both ends. The transmission line is simulated in Electromagnetic transients program (EMTP) and only fault currents are recorded at local end of the transmission line. Third-level WPD with mother wavelet db1 is utilized to calculate wavelet packet energies of fault current at each node. The normalized values of these features are fed to the RBF neural network to estimate fault location on both overhead section and an underground cable. The algorithm is fast as only half cycle post-fault data are sufficient and need not identify the fault sections (overhead line or underground cable) for estimating fault location. The accuracy of fault localization is very high irrespective of fault resistances, fault inception angles (FIA), and fault types at different locations on the hybrid transmission line.