Clarity-Optimized Wavelet With Autoencoder- ReliefF Ranking for Enhanced UHF PD Signal Feature Extraction

This article investigates advanced signal processing methodologies, with a focus on wavelet-based techniques, for the analysis of time-domain partial discharge (PD) signals captured using ultra-high frequency (UHF) sensors. The raw signals are systematically processed through a sequence of operations including bandpass filtering, wavelet-based denoising, DC offset removal, and pulse extraction. Each processing stage is critically examined in both time and frequency domains to ensure signal integrity and noise suppression. Emphasis is placed on the optimization of wavelet parameters alongside extraction of key temporal and amplitude-based features such as time difference, charge difference, pulse height, rise time, and pulse width. To address the challenge of identifying the most discriminative features, this work integrates advanced feature ranking algorithms, namely, auto-encoder-based ranking and the ReliefF method. Their effectiveness is evaluated through criteria including training convergence, reconstruction error, and relevant quality metrics. The methodological novelty lies in the systematic fusion of optimized wavelet-based signal conditioning with robust feature selection frameworks, enabling a comprehensive assessment of feature importance and clarity. A total of 7 wavelet transformations with various decomposition levels, 17 wavelet families with various sub-categories and 2 non-wavelets are involved in this study. Comparative analysis between wavelet-based and conventional non-wavelet methods demonstrates the superior performance of the former in terms of feature extraction fidelity and signal enhancement. The findings establish that the proposed clarity-importance ranking framework significantly advances the accuracy and efficiency of UHF PD signal processing. This contributes to enhanced interpretability and reliability in PD diagnostics, thereby supporting more effective monitoring and maintenance of high-voltage insulation systems in real-world applications.