Transient Power Quality Analysis Based On Wavelet Transform And Neural Network

This dissertation mainly researched on application of wavelet algorithm in signal detection and analysis of the transient power quality, combining artificial neural network technology to realize identification and classification of transient power quality. Five kinds of common transient power quality, namely voltage sag, voltage swell, voltage interruption, oscillatory transient and impulsive transient were main research objects. For the detection and analysis of power quality transient the following respects were researched.The definition, classification and research of power quality were introduced more comprehensively. The dissertation amply introduced the principles,application situation and existing problems of several transient power quality detection and analysis methods which were used frequently.For the denoising problem in actual signal detection, the improved soft threshold denoising algorithm based on wavelet packet transform was proposed. Different threshold method was used to process coefficients from different bands of wavelet packet decomposition. Fixed threshold was used to process low frequency coefficients and adaptive threshold based on Stein's unbiased risk estimating was used to process high frequency coefficients. Compared with the old wavelet packet method, simulation results showed that de-noising effectiveness was improved by this method.For the problem of compression in actual signal detection, a new compression algorithm based on second generation wavelet transform was proposed. A helpful attempt was done in order to do analysis transient power quality with second generation wavelet transform. The simulation results showed that high compression ratio can be obtained using the second generation wavelet transform but it has reconstruction error. In order to reduce reconstruction error under the condition of maintain a relatively high compression ratio, it requires further research.According to the non-steady characteristic of transient power quality, the singularity was extracted using wavelet packet decomposition after denoising and compression. The disturbance location of transient power quality was realized accurately and the characteristic indexes of signals can be obtained.Finally, the dissertation discussed the classification of transient power quality based on wavelet packet transform and neural network. Transient power quality signals were decomposed by wavelet packet transform.The energy of wavelet transform coefficients was extracted and normalization processed. Then the final eigenvector was formed and input to a neural network to construct neural network identification system for transient power quality. The simulation results verified the validity of this method.