摘要

Aiming at pattern recognition of on-line partial discharge (PD) monitoring, the wavelet packet coefficient matrix is constructed on the basis of wavelet packet decomposition of de-noised PD signal after the wavelet packet decomposition of de-noised partial discharge signal is done. Then, by the singular value decomposition of the wavelet packet coefficient matrix, the singular value energy percentage is defined as the feature vector of the partial discharge signal. Two classifications of the supportive vector machine are extended to multi one by M-ary algorithm, and the particle swarm optimization algorithm is used to optimize the parameters of supportive vector machine. Finally, input is regarded as the feature vectors, supportive vector machines are used to recognize 4 kinds of discharge signals, and a comparison of recognition effect is made by means of BP neural network. The results show that the feature vector of the singular value energy percentage can reflect the characteristics of the original signal well. Based on supportive vector machines, the discharge signals can be effectively identified with a 95% average recognition rate. And with the increase of decomposition scale, the average recognition rate of 4 kinds of discharge signal increases, but the increment decreases. Supportive vector machine and BP neural network can well identify 4 kinds of discharge signals, and the former has a better recognition effect.