AI-based synthetic data generation techniques for improved fault classification in power systems

The unfaltering action of electricity networks depends on the early identification of faults. Several machine learning (ML) and deep learning (DL) techniques have been applied for fault detection and classification. However, their performance is highly reliant on the quantity and excellence of the dataset. This paper proposes AI-based synthetic data generation techniques that provide an excellent option for producing synthetic data. This synthetic data can then be used in combination with the original dataset, to generate a new dataset. This strategy involves augmenting the current dataset with synthetically created dataset points, resulting in a new dataset. This new dataset leads to more robust and efficient fault identification and classification using either machine or deep learning techniques. The techniques train on a collection of actual datasets and produce an artificial dataset that mimics the features of the actual dataset. The results show that the new dataset with the original and synthetic dataset combined shows better fault classification results. It is also observed that the performance of the ML algorithms is reliant on the method used for creating the dataset. GAN and VAE-GAN emerged as the most adjustable methods, which enhanced the generalizability of different models. It was found that the Decision Tree and Gaussian Naïve Bayes models are the most effective, with near-perfect metrics across the majority of synthetic data generation techniques implemented in this paper.