Research On Classification And Identification Method Of Power Quality Disturbance Signal Based On Coordinate Transformation

The grid connection of new energy generation and the use of a large number of power electronic equipment make the problem of power quality disturbance more and more serious.The power quality disturbance problem will not only destroy the stable operation of the power system,but also cause serious safety accidents.Therefore,accurate classification and identification of power quality disturbances have always been the focus of scholars at home and abroad.Hilbert-Huang transform,wavelet transform and other methods are used for the detection and classification of power quality disturbances.However,because power quality disturbances often occur randomly in time and the amount of disturbance is difficult to capture,the detection and classification accuracy is not high.In this thesis,the power quality disturbance signal is converted into a complex plane curve by the coordinate transformation method,and a typical power quality disturbance classification model based on the complex plane curve image is established.The main work is as follows:1.Study the coordinate transformation method of power quality disturbance signal from Cartesian coordinate system to polar coordinate system.Aiming at the problem that the power quality disturbance signal curve in the Cartesian coordinate system is difficult to identify,In this thesis,the Hilbert transform is used to convert the disturbance signal from the Cartesian coordinate system to the polar coordinate system.N-fold rotational symmetry characteristics,research and prove the periodicity of the instantaneous amplitude,the periodicity and linearity of the instantaneous phase at the same time,and the composition of the complex plane curve can be analyzed by extending the two-dimensional complex plane curve to the three-dimensional space through Euler’s rotation theorem and time-shifted state.The instantaneous amplitude and instantaneous phase curves of the power quality disturbance signal in the complex plane of the polar coordinate system have a high degree of identification.2.Construct a complex plane curve image dataset of power quality disturbance signal.Power quality disturbances can be divided into single disturbances and compound disturbances.Considering that the types of compound disturbances cannot be exhausted,in this thesis,ideal sine,10 single disturbances(4 of which are single disturbances of different sub-harmonics),and 6 common compound disturbances,a total of 17 scenarios were analyzed.Since both the single and compound disturbance signal models contain amplitude control variables and time control variables,when one or two of them change,the two-dimensional complex plane curve of the disturbance signal will also change accordingly.In order to avoid image overfitting,this thesis sets the amplitude control variables and time control variables in the signal models of 17 situations as random numbers within the specified range when constructing the data set.In the case of noise-free and white noise signal-to-noise ratios of 20 d B,30 d B,40 d B,and 50 d B,respectively,calculate the two-dimensional complex plane curve images in 10 states of each disturbance signal and take 50 images each in total.In this thesis,the power quality disturbance signal complex plane curve image data set is divided into training set and test set according to the ratio of 4:1.The gray level cooccurrence matrix and the directional gradient histogram are selected as feature descriptors to describe the features of the images in the dataset,and the image features are fused into a feature vector according to the principle of equal weight.3.Construct a multi-classification support vector machine model of the complex plane curve image of the power quality disturbance signal.In this thesis,multiple twoclass classifiers of the vector machine model are combined in a one-to-one manner to realize the multi-classification problem of power quality disturbance.The constructed multi-class SVM model of complex plane curve image is composed of 136 two-class classifiers,and the prediction result is the class with the most votes for the multi-class vector machine model.Experiments show that the average class accuracy,precision,recall and F1 score of the trained complex plane curve image multi-class support vector machine model are over 99%,over 88%,over 75%,over 81%,and above 4 The macroaverage and micro-average results of the evaluation indicators are both above 95%.Compared with classification models such as K-nearest and decision tree,the multiclassification support vector machine model of complex plane curve image studied in this thesis has better classification and recognition results than K-nearest and decision tree model under the same conditions.