A Novel Adaptive Parameter Optimization Method for Denoising Partial Discharge Ultrasonic Signals

Denoising is a crucial step in ultrasonic monitoring of partial discharges (PDs) in transformers, as the signals are weak and susceptible to significant white noise interference in the field. Most denoising algorithms require manual parameter tuning, making it challenging to achieve effective field denoising and consistent denoising performance. To solve these problems, this article proposes an adaptive denoising algorithm. The algorithm is divided into two steps: preliminary denoising and secondary denoising. The preliminary denoising combines variational mode decomposition (VMD) with the Ljung-Box (LB) white noise test. The signal is decomposed into different intrinsic mode functions (IMFs), and the LB white noise test is used for the adaptive classification of PD components and noise components. After preliminary denoising, the signal-to-noise ratio (SNR) of the original signal is improved. After that, the particle swarm algorithm (PSO) is combined with wavelet threshold denoising (WTD) to perform secondary denoising of the signal. The algorithm improves the threshold value by adapting it to the field noise level and enhances the threshold function to resolve issues of instability and fixed deviation in traditional functions. Meanwhile, adjustment coefficients are introduced to them so that they can be optimized in the denoising process. A novel denoising evaluation metric, the Q(LB) of noise residual (QNR), which provides a more accurate assessment of denoising performance compared with traditional metrics is proposed. Taking it as the objective function, the PSO algorithm is used to achieve an adaptive optimal selection of parameters in the denoising process. The denoising results of both the simulated and experimental PD signals show that the proposed method can effectively remove the white noise and recover the PD signal more accurately.