A New Method Of Magnetizing Inrush Current Recognition Based On Wavelet Packet And Support Vector Machine

The presence of transformer magnetizing inrush current can cause misoperation of differential protection,thereby reducing the reliability of the protection action.Therefore,accurate and rapid identification between the magnetizing inrush current and fault current plays a crucial role in the stable operation of the transformer.However,the existing identification methods on magnetizing inrush current generally have the disadvantages of low accuracy or slow recognition.To this end,this paper proposes a new method of magnetizing inrush current identification based on wavelet packet and support vector machine.The specific implementation is to decompose the current signal based on wavelet packet decomposition to construct a new criterion for identification on current type,and then use the algorithm SVM for classification for the magnetizing inrush current and internal fault current in order to achieve the purpose of magnetizing inrush current identification.The main contents of this paper are as follows:1)Simulated the waveform of magnetizing inrush current and fault current in transformers using PSB toolbox in MATLAB/Simulink.In order to do some research on the influence of initial phase angle and residual magnetization on the identification of magnetizing inrush current,this paper simulates the magnetizing inrush current and fault current with different initial phase angles under three kinds of residual magnetic conditions.2)The waveform samples of the magnetizing inrush current and the fault current are decomposed based on the wavelet packet decomposition technique.Firstly,considering the performance of several common wavelet according to the regularity,orthogonality,symmetry and vanishing moment.Choose the Daubechies wavelet which is more suitable for this research.And divide the effect in term of the operating efficiency and frequency band'of Daubechies wavelet under different orders.Finally,selected 4th-order Daubechies wavelet(Db4)as the wavelet basis for decomposition of wavelet.Then,there are both advantages and disadvantages of the computation and decomposition accuracy on different decomposition scales,so the third level decomposition with less computation is chosen as the decomposition scale on the premise of decomposition accuracy.The following,the integrated energy calculation is performed on the waveform signals of the eight sub-bands on the third layer,and discuss the results of energy calculation under different integral exponents(?).Finally,choose 0.5 as value of ? because it can make the energy distribution of each sub-band more uniform.The energy feature vector is constructed based on the energy of the eight sub-bands as a new criterion for identification of current type.3)Based on the above work,this paper uses Support Vector Machines(SVM)as an intelligent classification algorithm to construct a current classification recognition model to accurately and quickly distinguish and identify the inrush current and fault current.First,the energy feature vectors of the simulated waveform are divided into training samples and test samples.Then,putting the energy feature vector of the training sample and the corresponding type of current into the SVM classifier for training,thereby obtaining a mapping relationship between the waveform energy feature vector and the current type.The SVM can classify the test sample according to the mapping relationship that has been acquired when the energy feature vector of the test sample is input into the SVM classifier that has been trained.Finally,the current type classification result of the SVM is compared with the actual current type of the test sample to obtain the accuracy of the SVM current identification.The results of the example show that:1)The average overall recognition accuracy of the three kinds of residual magnetic conditions is about 80%,and the average excitation inrush current identification preparation rate is about 95%.2)The samples with incorrect classification are mainly distributed between the single-phase ground fault and the three-phase winding fault in the zone.These two current belong to the internal fault current and have little effect on the magnetizing inrush current identification.3)The classification results under the three remanence conditions are basically the same,and different remanence will not affect the accuracy of the SVM recognition model.In summary,the new recognition model proposed in this paper is effective and feasible.