Transmission Line Fault Location Analysis Using Deep Network Designer

Abstract

Fault location in transmission lines is essential for maintaining power system reliability, reducing outage duration, and improving maintenance efficiency. Conventional fault-location methods, including impedance-based and traveling-wave approaches, often experience degraded performance under high fault resistance, waveform distortion, and noisy measurement conditions. To address these limitations, this study proposes a low-code deep learning framework for transmission-line fault location estimation using a long short-term memory (LSTM) network implemented in MATLAB Deep Network Designer (DND). A 100 km, 400 kV three-phase transmission line was modeled in MATLAB-Simulink, and fault cases were generated by varying fault types, locations, resistances, and inception angles. The resulting three-phase current signals were normalized and arranged as input sequences for regression-based fault-distance estimation. The dataset comprised 7,200 simulated cases per noise condition and was divided into training, validation, and testing subsets in an 80:10:10 ratio. The model was trained using the noise-free dataset and then evaluated under three signal conditions: noise-free, 30 dB additive white Gaussian noise (AWGN), and 20 dB AWGN, in order to assess cross-noise generalization. The proposed LSTM model achieved average RMSE values of 0.8575 km, 1.0000 km, and 1.1330 km, respectively, while maintaining R² values above 99.86% and an overall ±1 km accuracy of 89.44%. Comparative analysis against CNN and CNN-LSTM models indicates that the proposed approach provides competitive accuracy with lower implementation complexity in DND. These results show that the proposed LSTM-DND framework is effective for single-fault location estimation under simulated operating conditions, although further validation with mixed-noise and field-recorded data is still required.