Research On Denoising, Detection And Identification Of Transient Power Quality Signal

Electric energy is gradually becoming the broadest used and indispensable energy inmodern society. Power quality includes stable power quality and transient power quality.The research of transient power quality only just starts now. Therefore, the research oftransient power quality is becoming an important research subject in modern powersystem, and has important significance.Based on the research of both home and abroad, this thesis aims at transient powerquality problems, has done some researches on power quality signal de-noising,disturbance detection and disturbance identification. The main work and achievementsare as follows:1. The thesis explains the involved basic theories first, and five traditional transientpower quality disturbance signals are analyzed and studied, such as voltage sags, swells,interruption, oscillations and pulse.2. A wavelet package improved threshold de-noising method is put forward in thispaper to enhance the de-noising effect of transient power quality disturbance signals. Thenew de-noising method can overcome the discontinuity of the hard threshold and theoffset of the soft threshold. The simulation results of transient power quality disturbancesignal de-noising show that the new method can not only reduce the loss of usefulinformation while de-noising, but also effectively de-noise white noise under differentSNRs, The de-noising effect is superior to the soft and hard threshold function.3. The singularity of signal can be signified as wavelet transform modulus maximumbecause of the good time-frequency. But, the real signals will be disturbed by noise,which brings troubles to signal detection. Therefore, de-noising becomes very important.The simulation results show that after de-noising the detection effect is better.4. An improved neural network--genetic algorithm optimized bayesian regularization neural network is put forward in this paper to enhance the identificationeffect of power quality disturbance. And the wavelet packet-energy entropy is used toconstruct a feature vector. The improved method can overcome the local minimumproblem of traditional BP neural network, the over fitting and too much network nodes ofgenetic algorithm optimized BP neural network.5. Finally, comprehensively summarized the full paper, and put forward furtherresearch.