Considering the time-frequency characteristic of the partial discharge (PD) signals, the kind of improved wavelet neural network is constructed by principle of the temporal-scaling approach, and it is that using temporal-scaling domain of the wavelet basis function being chosen covers that of the partial discharge signals. It is fact that the PD signals are transformed by the wavelet based on the different scales and displacements, and the wavelet detail coefficients are obtained under the different scales, which are inputted into the wavelet neural network to accomplish pattern classification during the network learning course. The PD signals, which are simulated with different PD signal sources in the lab, are de-noised and normalized. Using the power spectrum analysis method, figure out the temporal-scaling domain of the PD signals and figure out that of the wavelet function chosen, till the two pairs of parameters can be conformed, then the seating factors and displacement factors of the wavelet function, used as the structure parameters of the wavelet neural network, are taken as the basic frame of the improved wavelet neural network. The results indicate that the improved wavelet neural network has not only better identifying ability than that of the BP neural network and pattern features of the PD signals could be automatically extracted, but also the recognition precisions are higher than that of the other networks
[1]
Harold H. Szu,et al.
Neural network adaptive wavelets for signal representation and classification
,
1992
.
[2]
Pauziah Mohd Arsad,et al.
Application of ANN to power system fault analysis
,
2002,
Student Conference on Research and Development.
[3]
Wu Wei-ling,et al.
An Improved Training Algorithm for Artificial Neural Networks
,
2004
.
[4]
Lin Hui.
A New Neural Network for Fault Diagnosis of Partial Discharge in Power Transformers
,
2005
.
[5]
Tan Ke.
Partial discharge recognition based on time domain features of pulse waveform
,
2001
.
[6]
Qinghua Zhang,et al.
Wavelet networks
,
1992,
IEEE Trans. Neural Networks.
[7]
Yagyensh C. Pati,et al.
Analysis and synthesis of feedforward neural networks using discrete affine wavelet transformations
,
1993,
IEEE Trans. Neural Networks.