Optimal Wavelet Packet Denoising Technique for the Detection of High-Impedance Faults in DC Microgrid

During data acquisition in dc microgrids (DCMGs), high-impedance fault (HiF) signals are often contaminated with various noise sources such as seismic interference, Gaussian or non-Gaussian noise, and power line interference. Effective preprocessing is essential to suppress these artifacts and extract reliable fault features. While wavelet packet denoising (WPD) has been used for HiF detection, its performance heavily depends on the choice of parameters, such as wavelet basis, decomposition level, rescaling, and particularly the thresholding strategy. To address this, this article introduces an adaptive thresholding technique, where the shape of the soft-thresholding function is dynamically tuned based on signal decomposition characteristics. The key contribution is to improve HiF detection by particle swarm optimization (PSO)-based tuning of wavelet parameters based on suitable selection of the basis function, thresholding, decomposition level, and rescaling. The proposed method’s efficiency is evaluated using root mean square error (RMSE) and signal-to-noise ratio (SNR). The case studies show that there is significant improvement in HiF detection using the proposed PSO optimized wavelet technique under broadband noisy scenarios between 20 and 60 dB and thus outperforms other techniques. The simulated results are also supported with the experimental validation in a constrained environment using the NVIDIA-JETSON NANO graphical processing unit (GPU) device before it can be upscaled to other relevant devices.