Extracting Partial Discharge Signals of Generator Based on Elevated Minimum Description Length Criterion and Fast Fourier Transform

A new method is put forward to remove the noise in partial discharge (PD) based on fast Fourier transform and elevated minimum description length (MDL) criterion. The main interference of generator PD signals is narrow-band interference and white noise. Firstly FFT is designed to deal with the PD signals to reduce the amplitude of narrow-band interference in frequency on basis of differences between them. Then mid-signals after FFT are decomposed by wavelet, the most appropriate wavelet coefficients are obtained by elevated MDL criterion, finally the corresponding coefficients are reconstructed to achieve de-noised PD. This method has wonderful virtues such as free from any parameters estimation about noise, free from threshold choosing behavior, so the method is highly adaptive. Large mount of simulation proves that this method can wipe off narrow-band interference and white noise efficiently, prior to the de-noising algorithm of wavelet and MDL, so it provides a new way for PD de-noising.