Abstract: Currently, partial discharge (PD) signals of high-voltage mining cables are easily buried in noise and difficult to extract. Variational Mode Decomposition (VMD) is an effective method for PD denoising, but the number of decomposition layers and penalty factor of the VMD algorithm are difficult to determine. To address this problem, a denoising method for PD signals of high-voltage mining cables based on Adaptive Spiral Flying Sparrow Search Algorithm (ASFSSA)-VMD-KSVD was proposed. ASFSSA was used to optimize VMD, and a chaotic mapping strategy was utilized to make the population distribution more uniform and avoid falling into local optima. A series of intrinsic mode functions (IMF) were obtained through VMD, and the Composite Multiscale Fuzzy Dispersion Entropy (CMFDE) was then used to screen the properties of IMF components, dividing them into signal-dominated components and noise-dominated components. For screened noise-dominated components, training samples were constructed for KSVD dictionary learning, and noise was further suppressed through sparse coding and dictionary updating. The processed coefficients were reconstructed, and the signal blocks were superimposed to obtain the denoised signal. The denoising performance was evaluated using Signal-to-Noise Ratio (SNR), Root Mean Square Error (RMSE), and Normalized Cross-Correlation (NCC). The experimental results showed that under different SNR conditions, the SNR after denoising using the ASFSSA algorithm was much higher than that of the GWO and IWOA algorithms, demonstrating a significant advantage in noise suppression. The RMSE after denoising using the ASFSSA algorithm was much smaller than that of the Grey Wolf Optimization (GWO) and Improved Whale Optimization Algorithm (IWOA) algorithms, indicating the smallest difference between the true and predicted values during denoising. The NCC after denoising using the ASFSSA algorithm was very close to 1, showing excellent waveform similarity.