Denoising Method for Partial Discharge Signals Based on SVMD_STNLM_GSTVD

ABSTRACT Currently, partial discharge (PD) signal denoising algorithms face issues such as complex parameter selection, incomplete noise suppression, and low operational efficiency. A hybrid denoising method (SVMD_STNLM_GSTVD) is proposed to address these issues, specifically integrating Sequential Variational Mode Decomposition (SVMD), Structure Tensor‐based Non‐Local Means (STNLM), and Group Sparse Total Variation Denoising (GSTVD). First, SVMD is used to adaptively decompose the original noisy PD signal, and then the correlation coefficient criterion is used to distinguish the modal components into high‐quality components dominated by effective signals and low‐quality components dominated by noise, achieving preliminary separation of signal and noise. For high‐quality components, in order to preserve signal structural features and effectively suppress noise, STNLM is introduced, and the chaotic sparrow search algorithm (CSSA) is used to intelligently optimise its key parameters. For low‐quality components, the GSTVD method is adopted, which introduces group sparse regularisation terms and total variational constraints to remove noise and avoid excessive smoothing of signal edge features. Finally, the SVMD_STNLM_GSTVD denoised signal is subjected to secondary filtering using a soft threshold function to further eliminate residual noise and improve the overall denoising effect. The simulation and actual measurement results show that the proposed method outperforms the compared methods in terms of signal‐to‐noise ratio, waveform similarity, root mean square error, and running time. It can effectively extract PD features and provide a reliable basis for subsequent fault diagnosis.