Application of neural network to suppress PD stochastic pulse noise for large generators

Abstract

In the continuous on-line partial discharge (PD) monitoring of large generators, the numerous interferences aroused during machine operation or coupled into the machine from the power system, so the noise cancellation techniques are essential to on-line PD measurement. According to the characteristics in time domain, the interferences can be divided into periodic narrow-band noise, the white noise, the periodic stochastic noise and the stochastic pulse noise. Among these interferences, the stochastic pulse noise is the most difficult to be eliminated because of the strong comparability in both frequency domain and time domain between the PD pulse and the noise pulse. The paper introduces the characteristic of difference PD signals and noises both in time domain and in frequency domain, proposing a new algorithm to suppress the stochastic pulse interference. The algorithm is based on the pattern recognition by the neural network (NN). The data sampled from both the on-line PD monitoring and off-line PD experiment over winding bar have been used to train and test the algorithm. The results given in the paper shows that the new algorithm is effective in suppressing of the stochastic pulse interference and the algorithm is useful to suppress other noises.