A deep learning-driven traveling wave method for GPS-free and noise-resilient fault location in compensated power networks

Abstract

This paper introduces a novel hybrid fault location technique for high-voltage transmission lines, integrating travelling wave (TW) principles, discrete wavelet transforms (DWT), and long short-term memory (LSTM) neural networks. The proposed method enhances fault detection speed, improves location accuracy, and demonstrates resilience against high-impedance faults. The LSTM network is specifically trained to detect the arrival of the initial wavefront through single-ended measurements, while DWT effectively extracts the high-frequency components of transient signals. A simulation of a 400 kV, 120 km transmission line, modeled on real parameters from the Algerian grid, was conducted using ATP-EMTP. The methodology was implemented in MATLAB and compared with several state-of-the-art approaches, including GPS-synchronized TW methods, under various noise conditions with signal-to-noise ratios (SNR) as low as 5 dB. Additionally, the influence of thyristor-controlled series compensators (TCSC) on location accuracy was explored. The results confirm the applicability of the proposed technique in modern wide-area protection schemes, especially for remote relays and next-generation digital fault recorders (DFRs).