Fault Location in Distribution Networks Based on SVGG16‐CBAM With Time‐Frequency Analysis

ABSTRACT

Accurate fault location in distribution networks is crucial for ensuring grid reliability and operational stability. This paper proposes a unified fault location framework based on a simplified VGG16 network integrated with a convolutional block attention module (SVGG16‐CBAM). The shared SVGG16‐CBAM backbone extracts discriminative fault features, while two task‐specific output structures perform fault section identification and exact fault point localization. The VGG16 network is simplified by reducing the number of fully connected layers to two, enhancing computational efficiency and model compactness, and CBAM emphasizes informative features while suppressing irrelevant ones to improve fault location accuracy. A 10 kV IEEE 13‐bus system was simulated in MATLAB/Simulink, and three‐phase voltage and current signals were transformed into time‐frequency graphs using continuous wavelet transform (CWT) to extract high‐level features. Experimental results demonstrate high accuracy for both tasks, and further verify the method's robustness, generalization ability, and interpretability through cross‐validation, robustness tests, unseen‐distance evaluations, and feature visualization with t‐distributed stochastic neighbor embedding (t‐SNE) and gradient‐weighted class activation mapping (Grad‐CAM).