Objective In the health monitoring of electrical equipment, partial discharge (PD) signals are often disturbed by various noise sources, which may come from the operational noise of the equipment itself or external environmental interference.

Method To effectively address the noise interference and improve the accuracy and reliability of PD detection, a denoising algorithm combining spectral analysis-based adaptive singular value decomposition (SVD) and empirical wavelet transform (EWT) is proposed. Initially, noisy PD signals undergo fast fourier transform (FFT) for spectral analysis, and an improved method combining classic threshold and spectral amplitude row vector kurtosis discrimination is proposed to determine the number of narrow-band interferences, reconstruct, and remove periodic narrow-band noise. Subsequently, the EWT algorithm is used to adaptively decompose the residual white noise in PD signals, selecting modal components that meet the kurtosis criteria to reconstruct the PD signals. Finally, an improved threshold method is utilized to remove the remaining small amounts of white noise, resulting in a denoised PD signal.

Result Simulation and actual denoising results show that the proposed method achieves improvements in signal-to-noise ratio, root mean square error, correlation coefficient, and noise reduction rate, reaching 7.02, 0.0112, 0.9003, and 33.0057 respectively.

Conclusion The method effectively removes narrow-band interference and white noise, and compared to other denoising methods, it shows improvements in multiple evaluation metrics, demonstrating good denoising performance.