Intelligent de-noising method for UHF partial discharge signals in power equipment based on hybrid wavelet-SVD algorithm

To improve the reliability of partial discharge (PD) monitoring in smart grid equipment, this paper proposes a hybrid denoising method for ultrahigh frequency (UHF) PD signals based on wavelet decomposition and singular value decomposition (SVD). First, the original UHF signal is decomposed using a five-level discrete wavelet transform (DWT) with the db4 wavelet. The Hilbert transform and moving average smoothing are then applied to the detail component with the highest energy to extract the signal envelope and accurately locate the PD event. Next, SVD is used to denoise the identified PD segment, with K-means clustering applied to classify singular values and determine the optimal number to retain. Compared with conventional methods such as DWT, empirical mode decomposition (EMD), and SVD-only approaches, the proposed method achieves a higher noise suppression ratio (NSR) while preserving the signal’s key features and spectral details. Experimental results from a 110 kV GIS field test demonstrate the method’s effectiveness in suppressing noise and enhancing PD signal clarity, making it well suited for high-fidelity diagnostics in intelligent power equipment monitoring systems.