An improved scale dependent wavelet selection for data denoising of partial discharge measurement

The detection of partial discharge (PD) signals has proven extremely important to diagnose the integrity of the insulation in high voltage equipments. The measurements of such signals are often accompanied by noise from different sources, which can compromise the data analysis. Numerous wavelet shrinkage denoising techniques have been discussed recently in the literature. This article proposes an alternative criterion for selection of the mother wavelet, named signal to noise ratio based wavelet selection (SNRBWS), where the wavelet selection is done, for each scale, based on the maximization of the signal to noise ratio (SNR) of the estimated signal. The detail or approximation coefficients corresponding to the largest peak amplitude value are seen as from the PD signal, while the ones corresponding to the lowest peak amplitude value are seen as from noise. Such coefficients are conceived through the decomposition of the PD signal previously obtained for a given equipment. The proposed method is compared to the energy based wavelet selection (EBWS) method and to the correlation based wavelet selection (CBWS) method for signals measured from current transformers, circuit breakers, generators, gas-insulated switchgears (GIS), surge arresters and transformers, and for simulated PD signals embedded in noise. The proposed method presented better denoising results when compared to the previously proposed methods for most of the tested signals, where among various assessment parameters analyzed were the correlation coefficient, the mean square error (MSE), and the SNR. The algorithm also showed superior performance compared to the others with respect to the processing time.