Spectral Kurtosis And Its Application In Power System Transient Signal Processing

As the size expansion of power grid and its increasing complex degree, quantities of transient signals are produced in the real operation of power system, because of occurred fault incidents, the operation of lots of power electronic equipment and bad power changes. Analyzing power system transient signals could provide useful information for power system early fault diagnosis and predicting the stable operation or failure condition of power grid, transmission lines and large power equipments, and could guide the repair works of power grid and power equipments. Therefore, analyzing power system transient signals is very important significance.A totally new transient signal analytical tool——spectral kurtosis method is proposed in this paper, which is used to pick up partial discharge signals and classify the transient power disturbance signals. The main research work in this paper is summarized as followed:1. Spectral kurtosis methods based on sectioning、short time Fourier transform、wavelet transform and Wigner Ville distribution are used to analyze a simulation signal contained three different kinds of components. These methods’ main properties and their abilities to characterize different signal components are discussed, and their main problems and some influence factors are researched in this paper.2. Partial discharge signals are frequently submerged in some noises. In order to extract the partial discharge signals, a new method based on the spectral kurtosis is proposed in the paper. Firstly, the spectral kurtosis of signals with noise is calculated based on STFT. Then according to relationship between the spectral kurtosis and wiener filter can design an adaptive optimal band-pass filter to filtering the signal with noise. Finally, ideal partial discharge signals features are acquired with wavelet smooth denoising. By suppressing smooth random interferences and narrow band period interferences, comparing the method in this paper with direct wavelet threshold denoising method. Various performance indicators are better than the wavelet threshold denoising can be reached.3. According to the inherent characteristics of the transient power quality disturbances and the characteristics of spectral kurtosis, a new method based on BWD of calculating spectral kurtosis which is combined with support vector machine to apply in transient impulse and oscillation disturbances classification is proposed in this paper. BWD spectral kurtosis method is used to calculate spectral kurtosis of transient pulses and transient oscillation which are two kinds of disturbed signals in the algorithm. The maximum, minimum and average of spectral kurtosis are chosen as characteristics, and then input into SVM for training and forecasting, whose parameters are optimized with PSO. Simulation data is got by PSCAD/EMTDC and analyzed with this method. The results show that the method based on BWD spectral kurtosis can effectively extract disturbance characteristics and has a good anti-noise capability; SVM classifier can effectively identify two kinds of disturbance and has a higher recognition rate for small samples and other disturbances superposition.