A Classification-Guided Adaptive VMD-Wavelet Denoising Algorithm for Cable Partial Discharge Signal Processing

In the context of strong electromagnetic interference, cable partial discharge (PD) signals typically exhibit a low signal-to-noise ratio (SNR), resulting in the masking of signal features by noises. Due to limitations such as fixed parameters and poor generalization across different discharge types, existing denoising methods cannot effectively suppress noise. This paper proposes an adaptive denoising approach integrating signal classification with dynamic parameter optimization. First, the short-time Fourier transform (STFT) combined with Oxford threshold segmentation is employed for time-frequency preprocessing, thereby enhancing the separability of impulse features under low SNR conditions. Next, a convolutional neural network (CNN) model is constructed to accurately classify the preprocessed time-frequency images, identifying the PD signal in three types (single exponential decay, double exponential decay, and double exponential decaying oscillation). Finally, based on the classification results, optimal variational mode decomposition (VMD) parameters (number of modes K and penalty factor α) and wavelet threshold denoising parameters (wavelet basis and decomposition levels) are dynamically allocated for different PD signal types, enabling adaptive VMD-wavelet joint denoising. Experimental results demonstrate that the proposed CNN-based classification model achieves an accuracy over 99% in complex and noisy environments; compared with traditional fixed-parameter VMD-wavelet methods, the adaptive denoising algorithm significantly improves denoising performance. The average SNR improvement exceeds 7 dB, the root mean square error (RMSE) is reduced by more than 15%, the normalized cross-correlation coefficient (NCC) approaches 0.1, and the algorithm can effectively reduce the time delay error in subsequent PD localization. This method substantially enhances feature extraction and localization accuracy of cable PD signals under strong noise interference, providing more reliable technical support for condition monitoring of power equipment.