Partial Discharge Signal Noise Suppression Using Spectrum Clustering and SVD-EEMD
Jingcong Gou, Zhongdong Wu, Bingkun Gan, Pengbo Wang · 2025
To tackle the problem of partial discharge (PD) signals from power transformers being overwhelmed by Gaussian white noise and narrowband interference, the accuracy of signal classification and localization, a novel noise suppression approach is proposed that integrates spectrum clustering with singular value decomposition (SVD) and ensemble empirical mode decomposition (EEMD). First, K-means clustering is applied to identify periodic narrowband interference, and the corresponding frequency amplitudes are set to zero. Next, a Hankel matrix is created from the noisy signal, followed by SVD decomposition. By truncating small singular values and performing diagonal averaging, the effective signal is reconstructed. In the final step, EEMD, along with Pearson correlation coefficient (PCC), PCC is applied to further mitigate noise in the PD signal. The experimental findings demonstrate that this approach effectively eliminates noise from the signal.