Research On Detection And Identification Of Power Quality Disturbance Based On Modified S-Transform

With the rapid development of smart power grid,the nonlinear power electronic load and sensitive equipment load are increasing day by day,which makes the power quality problem in the power grid worse and worse.Due to the strong randomness of each power quality disturbance signal in the power grid,it is difficult to detect,and it is difficult to find the exact cause of the accident,which is vividly known as the ghost phenomenon of the power system.It not only affects the normal operation of the power equipment,but also causes more serious losses in the power grid and poses a major security hazard to the power supply system.In order to ensure the safe and stable operation of power system and reduce the loss of national economy,the fast and accurate detection and identification of power quality disturbance signal has become the top priority in power quality research.Therefore,this thesis proposes corresponding improved algorithms around power quality disturbance detection and recognition,the main research contents are as follows:(1)Aiming at the problem that the detection accuracy is not ideal and the existing detection methods can not satisfy the accurate detection of a large number of complex disturbances when detecting non-stationary composite power quality disturbance in power grid,a new method to detect the disturbance is proposed based on adaptive modified incomplete S-transform.By introducing adaptive window adjustment factor to adjust the scale of Gauss window,the time-frequency resolution of different power quality disturbance can be satisfied.Firstly,the multi-scale maximum algorithm is used to extract the feature frequencies of the multiple power quality disturbance events.Then by defining the residual difference curve to amplify the power quality disturbance mutation local part,the number of pulses can be obtained and this is used as a basis to combine the feature frequencies so as to adaptive determine the window adjustment factor.Finally,the feature frequencies are transformed locally to realize the adaptive detection of the signal.The simulation experiment and measured signal analysis show that the proposed method has strong anti-noise,low computational complexity and high detection accuracy,and is suitable for time-frequency analysis of power quality disturbance signals.(2)Aiming at the problem of power quality disturbance signal classification and recognition,a method of power quality disturbance event recognition based on 2D fast discrete orthogonal S transform and metric learning is proposed.Since the power quality disturbance are essentially 1D signals,the traditional power quality recognition method is to find the feature extraction and classification recognition method based on 1D signal processing method.This thesis proposes to use 2D signal processing to extract features and convert 1D signals into 2D signals with the same number of rows and columns.Firstly,based on the mathematical modeling,17 different kinds of noise signals are synthesized and transformed into 2D signals with equal rows and columns.Secondly,the amplitude matrix is obtained from the 2D signal by 2D fast discrete orthogonal S transform,and then the features based on statistics and energy and image are created.Then,using the measurement theory,an optimal measure is learned by the KISSME method to reduce the correlation between the sample feature vectors.By using SVM classifier to construct a power quality disturbance classification model.The simulation results show that the proposed algorithm can effectively improve the accuracy of classification algorithm and the robustness to noise.