A Denoising Algorithm for Cable Partial Discharge Signals Based on Image Information Entropy and Multivariate Variational Mode Decomposition

In recent years, cross-linked polyethylene cables have been developed in transmission lines and urban distribution networks due to their advantages, lightweight, high-temperature resistance, and high transmission power. With the increase in the number of XLPE power cables put into operation and the extension of cable lines, silicone rubber insulated prefabricated intermediate joints are widely used in XLPE power cables due to their excellent high-voltage shielding performance, overall prefabricated design, reliable grounding, large voltage margin, and convenient on-site installation. The intermediate cable joint is usually installed and formed on the laying site, which can quickly leave a hidden danger of cable operation failure. Because it comprises multilayer solid composite structures with different dielectric properties, the probability of accidents is much higher than that of the cable body. Partial discharge (PD) detection is the main means of evaluating the insulation status of XLPE cables and the manufacturing and installation defects of cables. A PD denoising method based on image information entropy and novel adaptive multivariate variational mode decomposition (MVMD) is proposed to address the issues of white noise, periodic narrowband interference, and poor adaptability in on-site detection of PD at cable terminals and intermediate joints. Firstly, optimize the parameters of the MVMD algorithm by integrating multiple factors, and then, based on parameter optimization, perform modal decomposition on the noisy PD signal. Secondly, the kurtosis of each eigenmode component is calculated, and the kurtosis of the sine signal and double exponential decay signal at the signal-to-noise ratio of 0dB is calculated by using the characteristic that kurtosis is sensitive to noise to distinguish the PD characteristic information from the noise interference component. Then, the 3σ criterion is used to filter white noise with normal distribution. Finally, based on the improved new wavelet threshold function, the reconstructed PD signal is denoised to obtain the denoised PD signal. The following conclusions can be drawn by comparing the method with other denoising algorithms: (1) The Spearman variational mode decomposition (S-VMD) can improve modal aliasing, but there is still residual noise in the denoised signal. Hence, the denoising effect is not ideal. (2) The novel adaptive ensemble empirical mode decomposition (NAEEMD) cannot wholly eliminate modal aliasing, resulting in a certain degree of displacement of the discharge starting position and affecting subsequent diagnosis and positioning. (3) Although the short-time Fourier transform and matrix factorization (STFT-SVD) can effectively suppress white noise and periodic narrowband interference, the denoised PD signal contains residual noise, and the execution efficiency of this algorithm is low. (4) By calculating various evaluation indicators, the method has a good denoising effect on the on-site noisy PD signal. At the same time, this method has the advantages of less time consumption and high execution efficiency. The following conclusions can be drawn: (1) The information entropy is used to measure the aggregation characteristics of the gray image distribution accurately and then to determine the certainty of the PD pulse signal. By constructing the information entropy of grayscale, the mode aliasing phenomenon of empirical mode decomposition (EMD), and other algorithms is overcome, and the accurate decomposition of noisy PD signals can be achieved, thus achieving accurate feature extraction. (2) Distinguish PD features from noise interference by calculating kurtosis values. Using the characteristic that kurtosis is sensitive to noise, the kurtosis value of the sine signal and double exponential decay signal at SNR=0 dB is calculated to accurately distinguish PD feature information and noise interference component, which lays a foundation for improving the denoising effect of PD signal. At the same time, filtering out noise interference components largely compresses data, reducing algorithm time consumption and improving execution efficiency. (3) The denoising effect of noisy PD signals on site has been improved by improving the wavelet threshold function and the threshold value. Combining the number of wavelet decomposition layers with the general threshold setting, a new type of wavelet threshold function with exponential decay is constructed, which corresponds to the mathematical model of XLPE cable PD signal, and then the detail coefficients of each layer after wavelet decomposition are accurately obtained, to improve the denoising effect of on-site noisy PD signals. © 2024 China Machine Press. All rights reserved.