Effective Denoising of Multi-Source Partial Discharge Signals via an Improved Power Spectrum Segmentation Method Based on Normalized Spectral Kurtosis

In the field of partial discharge (PD) analysis, traditional methods typically employ single-source PD signal-processing techniques. However, these approaches exhibit significant limitations when applied to transformers with relatively complex structures. To overcome these limitations and achieve precise characterization of composite PD signatures, this study proposes an improved power spectrum segmentation method (IPSK) based on spectral kurtosis. Firstly, normalized power spectral kurtosis is used to select the appropriate parameters. Then, through the improved power spectrum segmentation method, the segmentation frequency band with the least noise is obtained. Finally, the instantaneous signal components with physical significance are obtained by reconstructing each frequency band through inverse fast Fourier transform. By analyzing the simulated signals and measured data of partial discharge, the proposed method is compared with EWT, AEFD, VMD, and CEEMDAN. The results show that IPSK has a good suppression effect on noise interference.