An Enhanced Data‐Driven Fault Detection Scheme in Power Systems Through Synthetic Samples Augmentation

This paper proposes an enhanced data‐driven fault detection scheme in power systems with sparse sensors. This scheme utilizes data augmentation techniques specifically designed to improve the accuracy of deep learning‐based fault detection classifiers. The enhancement is achieved by expanding limited training data through the implementation of the Auxiliary Classifier Wasserstein Generative Adversarial Network with gradient penalty (ACWGAN‐gp) network. The three‐phase voltage signals from sparse sensor locations in power grids are converted into grayscale images, simplifying the application of image classification for fault detection. A two‐stage Convolutional Neural Network (CNN)‐based fault detection scheme first determines a potential faulted subnetwork and then identifies the faulted line within it. The ACWGAN‐gp‐driven data augmentation generates high‐quality synthetic samples that are not duplicates but varied representations, enriching the dataset in quantity and fault scenarios. The effectiveness of this approach is demonstrated through tests on the IEEE 39‐bus test system.