Fault Diagnosis Of NPC Three-level Inverter Based On Support Vector Machine Optimization

With the development of new energy technology,solar energy,wind energy,tidal energy and other new energy are more and more popular in people's lives.As an indispensable component in the development of new energy,inverter is also more and more concerned by people.NPC three-level inverter is widely used in high-speed rail,electric vehicle and photovoltaic power generation because of its low harmonic,high efficiency,small size,light weight and high power density.However,compared with the two-level inverter,the three-level inverter uses more power tubes.Each of which carries high voltage,high current and high temperature,which is easy to cause power tube failure under frequent switching operation.In the inverter fault,about 60% of the faults are caused by the open circuit fault of the power tube.After the open circuit fault of the three-level inverter,the fault characteristics are not obvious,and the equipment can still operate,but it works in an abnormal state.Long time in this abnormal operation fault state will lead to equipment damage or even accidents.In order to obtain the open circuit faulted state of three-level inverter in real time and ensure the stable and safe operation of the application system,this thesis studies the open circuit fault diagnosis of NPC three-level inverter.Relevant innovative research has been done in the following three aspects:(1)A method of fault feature extraction based on energy proportion of ensemble empirical mode decomposition(EEMD)and kernel principal component analysis(KPCA)is proposed.In this method,EEMD is used to decompose the fault signal,the energy proportion of each intrinsic mode function(IMF)and residual term is used as the fault feature,and the dimension of KPCA feature is reduced,so as to eliminate redundant signal and obtain the fault feature with lower dimension.(2)A classifier optimization method of KNN-SVM is proposed.This method first optimizes the Gaussian kernel width parameter to obtain the best projection direction of kernel Fisher,then uses the projection of samples of this direction to determine the interval of KNN reference point,and finally uses the interval to determine the test samples for SVM and KNN classification and test them.The two algorithms complement each other to improve the accuracy of fault diagnosis.(3)An open circuit faults diagnosis method of three-level inverter based on KNNSVM optimal classifier is proposed.Firstly,the combination of EEMD and KPCA is used to extract the open circuit fault feature of three-level inverter,and then the fault feature is input into KNN-SVM classifier to realize fault diagnosis.Simulation results show that the feature extraction method based on EEMD-KPCA has a good effect.Compared with the combination of traditional wavelet and energy,the feature extracted by this method has the characteristics of high evaluation index and low feature dimension.The classifier optimization method based on KNN-SVM can effectively improve the test accuracy near the classification hyper plane compared with the traditional SVM and DT-SVM In order to improve the overall classification accuracy,the three-level inverter fault diagnosis method based on KNN-SVM optimization classifier,the classification accuracy and the overall classification accuracy near the classification hyperplane are better than the traditional SVM and DTSVM.Compared with BPNN and elm,it has higher advantages in diagnosis accuracy and test time,and better in robustness.