Efficient automated detection of power quality disturbances using nonsubsampled contourlet transform & PCA-SVM

The increasing complexity of modern power systems, driven by smart grid technologies, dispersed generation, and renewable energy sources, has made power quality (PQ) evaluation more challenging. Overuse of sensors and large data volumes further complicate accurate PQ assessment. This paper proposes a flexible signal decomposition methodology using the nonsubsampled contourlet transform (NSCT) for accurate PQ event detection. The NSCT's multiscale, multidirectional, and shift-invariant properties enable the decomposition of signals into transient and oscillatory components. High-frequency NSCT subbands are fused to extract oscillatory portions, while low-frequency subbands are averaged to detect transients. Morphological component analysis (MCA) and the split-augmented Lagrangian shrinkage algorithm (SALSA) optimize this process. Principal component analysis (PCA) is applied to the extracted features to reduce dimensionality and improve feature separability. These optimized features are used for training a multi-class support vector machine, with its parameters further optimized for enhanced classification accuracy. The proposed approach demonstrates superior frequency selectivity, adaptability, and computational efficiency, making it suitable for robust PQ disturbance identification and a wide range of signal-processing tasks.