Classifying Power Quality Issues in Railway Electrification Systems Using a Nonsubsampled Contourlet Transform Approach

Abstract

Railway electrification systems (ESs) pose significant challenges due to highly variable power demands and dynamic train operations. Effective power quality (PQ) monitoring for high-speed trains (HSTs) is crucial for maintaining stable and uninterrupted system performance. Rapid changes in load profiles can result in voltage sags, swells, frequency deviations, and harmonic distortion. Traditional metering systems often face limitations in accuracy and communication reliability under these conditions. This study proposes a robust signal decomposition and classification framework combining nonsubsampled contourlet transform (NSCT) with morphological component analysis (MCA) to accurately identify PQ disturbances. NSCT's shift-invariant, multiscale, and multidirectional capabilities allow for precise separation of oscillatory and transient components, while the split augmented Lagrangian shrinkage algorithm enhances decomposition efficiency. Features such as signal energy, entropy, and trend energy were extracted and visualized in a 3D feature space, demonstrating clear clustering for different PQ events. The system was tested using synthetic and Kaggle-derived datasets, achieving a classification accuracy of 100%, precision of 99.6%, recall of 99.3%, and F1-score of 99.4% across five event classes: Normal, Sag, Swell, Harmonic, and Noise. The results validate the NSCT-MCA framework's capability to reliably detect and distinguish PQ disturbances under noisy and fluctuating railway conditions, reinforcing its suitability for real-time deployment in modern electrification infrastructures.