Abstract
This paper presents a novel multiscale signal processing framework for power quality disturbance (PQD) and cyber intrusion detection in smart grids, combining Non-Subsampled Contourlet Transform (NSCT), Split Augmented Lagrangian Shrinkage Algorithm (SALSA), and Morphological Component Analysis (MCA). A key innovation lies in an adaptive weighting mechanism within NSCT's directional sub bands, enabling dynamic energy redistribution and enhanced representation of both low-frequency anomalies (e.g., voltage sags/swells) and high-frequency distortions (e.g., harmonics, transients). SALSA-based sparse optimization achieves an average signal-to-noise ratio (SNR) improvement of 12.8 dB, preserving essential transient structures, while MCA isolates fault-relevant morphological components for better interpretability. Extensive simulations on both synthetic signals and the IEEE 14-bus test system demonstrate detection accuracies of 98.6% for PQDs and 97.2% for cyber intrusions, including False Data Injection (FDI), Denial of Service (DoS), and Command Injection attacks. Each intrusion exhibits unique time-frequency scalogram signatures, which are effectively visualized using high-resolution, denoised 2D/3D spectrograms generated via adaptive Q-Factor Wavelet Transform (AQWT) and Short-Time Fourier Transform (STFT). Compared to baseline methods like STFT-only and DWT-SVM pipelines, the proposed NSCT-SALSA-MCA framework improves detection precision by 14-18%, reduces false positives by 22%, and remains robust under 30 dB noise and 20% data loss. Incorporating AI-driven anomaly detection and resilient state estimation further enables early flagging of compromised measurements, securing applications such as Economic Dispatch and Optimal Power Flow (OPF). The resulting scalograms provide interpretable visual insights, marking a significant advancement in smart grid monitoring with potential for real-time deployment at the edge.