Classification of Power Quality Disturbance with Synchrosqueezing and Vision Transformer

The rapid expansion of renewable energy sources (RES) in modern power systems has increased the occurrence of power quality disturbances (PQDs), affecting system performance. This paper introduces a novel approach combining Fourier Synchrosqueezing Transform (FSST) for time-frequency representation (TFR) with Vision Transformer (ViT) for classifying PQDs. By incorporating global characteristics, the ViT model increases classification accuracy, while GradCAM visualization improves interpretability. Experiments on synthetic datasets with single-type signal disturbances validate the proposed approach, showcasing superior performance in metrics compared to traditional methods. With a 99.73% accuracy rate, the proposed approach shows promise for precise classification and efficient power quality event monitoring.