Classifying Power Quality Disturbances Using Vision Transformers and SPWVD Images

What is it about?

Modern power grids are increasingly affected by renewable energy integration and power-electronic loads, which can introduce power quality disturbances (PQDs) such as sags, swells, harmonics, and transients. These disturbances can disrupt sensitive equipment and reduce the reliability and efficiency of smart grids. Many existing PQD classification methods struggle because PQDs are often nonlinear and non-stationary, making them difficult to represent and classify consistently. This work introduces a two-step approach that converts PQD signals into informative time–frequency images and then classifies them using a transformer-based model. First, PQD waveforms (generated in MATLAB following IEEE 1159 guidance) are transformed from 1-D signals into 2-D time–frequency representations using the Smoothed Pseudo Wigner–Ville Distribution (SPWVD), which provides rich feature detail. Second, these images are fed into a Vision Transformer (ViT) classifier that uses self-attention to capture global patterns across the time–frequency plane. The proposed ViT–SPWVD method achieved 98.94% classification accuracy, demonstrating strong potential for robust PQD detection and classification. To our knowledge, this is among the first applications of a vision transformer to PQD analysis, suggesting a promising direction for transformer-based monitoring in power systems.

Featured Image

Photo by Mohammad Mardani on Unsplash

Photo by Mohammad Mardani on Unsplash

Why is it important?

Modern power grids increasingly experience “electricity glitches” (power quality disturbances) caused by renewables energy integration and practical loads, and these glitches can disrupt sensitive equipment, trigger shutdowns, or shorten asset life—so detecting and correctly identifying the disturbance type quickly is crucial. Our approach turns raw electrical waveforms into clear time-frequency “images” (using SPWVD) and then uses a Vision Transformer to recognize the disturbance patterns more reliably, enabling smarter, faster monitoring and troubleshooting that can reduce downtime, protect equipment, and improve grid reliability.