Voltage Sag Data Generation Based on the Probabilistic Diffusion Model

Abstract

To effectively analyze and mitigate voltage sag phenomena using deep learning techniques, a diverse and representative dataset of voltage sag events is essential. However, the available voltage sag dataset obtained from monitoring systems is often limited in quantity and lacks comprehensive coverage of various sag types. In this paper, the achievement of generating high-quality voltage sag waveforms that closely resemble real-world scenarios is accomplished by leveraging the diffusion model’s ability to capture complex data distributions. Our study focuses on modifying the diffusion model to capture the specific characteristics of voltage sag waveforms. The results demonstrate the effectiveness of the proposed approach in generating realistic voltage sag datasets, highlighting its potential for various applications in power system analysis and planning.