Collaborative localization of WGAN-GP and edge federated learning for single-phase grounding faults in distribution networks

Abstract With the rapid development of new power systems, medium- and low-voltage distribution networks have become increasingly complex. This leads to low accuracy and poor generalization in traditional localization methods, which cannot meet the requirements for rapid self-healing and safe operation in intelligent distribution networks. To overcome the drawbacks of existing fault localization techniques—such as limited fault data, inadequate extraction of temporal information, multi-source data heterogeneity, and the need for lightweight models—a novel collaborative fault localization approach based on Wasserstein Generative Adversarial Network with Gradient Penalty and edge federated learning is proposed. The improved Wasserstein Generative Adversarial Network is used to generate high-quality fault samples and extract temporal features. Adaptive weighting and a region-weighted module are introduced into the Federated Averaging algorithm to improve global aggregation accuracy. This study utilizes the IEEE 123 simulation dataset and the on-site measured dataset as verification data, which consist of over 10,000 simulation electrical samples and 326 real single-phase grounding fault samples, respectively. According to the results, the Fréchet Inception Distance is 12.36 for fault sample generation, while the Peak Signal-to-Noise Ratio reaches 27.94 dB. The proposed method achieves a localization accuracy of 94.85%. In cases of high resistance, arc faults, load variation, and noise, the localization accuracies reach 82.70%, 84.06%, 89.37%, and 82.25%, respectively. The proposed model also performs better than existing methods in terms of parameter size, inference time, and bandwidth consumption. The proposed model outperforms existing methods in terms of fault sample generation, localization accuracy, anti-interference capability, and lightweight performance. Moreover, it does not require the uploading of raw measurement data, thereby balancing data privacy protection with the deployment capability of edge terminals. It is highly suitable for complex on-site operating conditions and provides technical support for rapid and accurate fault localization, on-site troubleshooting, and self-healing in medium- and low-voltage distribution networks.