Multi-source partial discharge interference suppression and signal separation for fault diagnosis in HVDC gas-insulated lines

Noisy mixed multi-source partial discharge (PD) signals pose a significant threat to the safe operation of high-voltage direct current (HVDC) gas-insulated lines (GILs). Most existing methods focus on denoising single-type PD signals and specific types of interference, often relying on prior assumptions about noise characteristics or the number of discharge sources. These methods struggle with realistic conditions involving multiple discharge sources, various noise types, multi-band PD spectra, and overlapping narrowband interference. To address these challenges, this paper proposes a fault diagnosis framework integrating interference suppression and signal separation. First, narrowband noise is suppressed in the frequency domain using a moving median threshold and Akima cubic interpolation, while white noise is reduced through a sign-error least mean square adaptive filter. Then, time-domain, frequency-domain, and time-frequency features are extracted from the denoised signals and standardized via Z-score normalization. The standardized features are first reduced using t-distributed stochastic neighbor embedding and then clustered using density-based spatial clustering of applications with noise to achieve signal separation. The proposed noise suppression and signal separation techniques are compared with existing methods, demonstrating superior performance. Experimental results under both controlled and near-realistic conditions confirm the proposed method’s effectiveness in suppressing narrowband and white noise and in successfully isolating multiple PD sources and external corona interference, providing a practical solution for intelligent fault diagnosis in HVDC GILs.