Partial Discharge Denoising of High-Voltage Cables for High-Speed Trains Based on Singular Spectrum Analysis and ICEEMDAN Decomposition

Abstract

Partial discharge (PD) detection is an effective method to evaluate the insulation status of cables for high-speed trains. Various factors in the detection process due to external interference make it difficult to get the ideal PD signal. Periodic narrow-band interference and random white noise are the main interference factors in the PD signal. A PD denoising method based on singular spectrum analysis (SSA) and improved adaptive noise-complete ensemble empirical modal decomposition (ICEEMDAN) is proposed in this article. With the method, by selecting the maximum value of singularity slope as the dividing point between narrowband interference and valid signal after grouping, the SSA algorithm selects the valid signal part for signal reconstruction, the noisy PD signal with white noise interference is decomposed into multiple eigenmodes by ICEEMDAN decomposition, which effectively avoids the modal blending in the empirical modal decomposition, sorts the intrinsic mode function (IMF) components further by the kurtosis criterion, and reconstructs the filtered IMF components, which extracts the pure ideal PD signal. Through denoising analysis of simulated PD signals and measured signals, the noise rejection ratio (NRR) of this method is 19.1836 in the laboratory and 16.389 in the depot, both higher than other methods. Therefore, this method has better denoising effect on noisy PD signals and can retain the original PD information to a higher degree.