Harmonic Fingerprinting for Power Quality Source Identification Using Non-Subsampled Contourlet Transform

The reliable operation of modern electrical grids requires precise identification of power quality disturbances caused by various sources such as non-linear loads, faults, and electric vehicle (EV) chargers. This study introduces an advanced approach for harmonic source identification using the Non-Sub sampled Contourlet Transform (NSCT), a method known for its superior time-frequency resolution and ability to capture multi-scale, directional features. By applying NSCT to voltage and current signals, we extract unique harmonic patterns that effectively distinguish between disturbance sources, including high-order harmonics from non-linear loads, low-frequency distortions from faults, and switching harmonics from EV chargers. To improve classification accuracy, a dual-task Convolutional Neural Network (CNN) is used, addressing both the classification of power quality events and the identification of fault locations. The results show that the proposed NSCT-based method, combined with the dual-task CNN, outperforms traditional Fourier-based techniques, offering faster convergence, higher classification accuracy, and minimal loss. This integrated approach provides a powerful tool for real-time power quality monitoring, enhancing grid reliability and operational efficiency while enabling accurate source identification in complex power systems.