Intelligent Identification and Localization Method for Cable Fault Types Based on Deep Residual Network and Reflection Method

Le résumé fourni par la source

Cable faults pose significant risks to power system reliability and safety. This paper proposes an intelligent method for cable fault type identification and precise localization by integrating deep residual networks (ResNet-50) with time-domain reflectometry (TDR) reflection signals. The approach processes raw or preprocessed reflection waveforms through a 50-layer ResNet architecture to extract hierarchical features, enabling simultaneous multi-class fault classification (normal, short-circuit, open-circuit, low-resistance ground, high-resistance ground) and regression-based distance estimation. Experimental results on a dataset of 5,000 simulated TDR signals (with SNR 20–40 dB) and supplementary real measurements demonstrate that the proposed method achieves 96.5% fault type identification accuracy and average localization error below 1 meter across fault distances up to 1000 m, substantially outperforming conventional TDR, SVM, standard CNN, and LSTM baselines. The method exhibits strong robustness to noise and offers a promising solution for automated, intelligent cable fault diagnosis in power distribution networks.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.