The Research On Fault Diagnosis Method Of Power Electronic Converter Circuits Based On Ensemble Classifier

With the rapid development of power electronic technology,the power electronic converter circuit as the core of power transform has been applied to industry,energy,transportation and other fields widely.However,the resulting converter circuit failure problems also increasingly highlighted,it will cause serious consequences once the failure happens.Especially open circuit fault of the device,electronic equipment and personal safety will be under serious threat if the fault is not been diagnosed and recovered in time.Therefore,study on diagnosis problem of power electronic converter open circuit fault deeply,and put forward the feasible solutions,which has great practical value and realistic significance within Intelligent,efficient and accurate diagnosis.There are some shortcomings in view of the existing fault diagnosis methods in nonlinear,multi-class classification,this paper choose three phase six pulse rectifier circuit as the research object.On the basis of the classical classifier,to put forward two converter circuit fault diagnosis methods by combining the ensemble learning technology and the classifier,using the simulation data and experimental data for validation.In this paper,the concrete research content is as follows:(1)Realized the simulation of thyristor open circuit fault and the fault signal extraction.Through cut off pulse to simulate the open circuit fault,22 kinds of fault waveform have been extracted in the dc output;sampling data from the 22 kinds of fault waveform,get the classifier training data and testing data.(2)A novel method is presented for converter circuit failure diagnosis based on BP-Adaboost strong classifier.The single BP neural network as a weak classifier,Adaboost algorithm is used to iterative and optimize the weak classifier,a strong classifier is composed of five weighted weak classifier.Putting the test data into the strong classifier,the final diagnosis is got by weak classifier with each weight.Results show that Relative to the single BP neural network classifier,the BP-Adaboost classifier in the fault diagnosis of power electronic converter circuit is reduced into the Over-fitting,improved the correct diagnostic rate.(3)A novel method is presented for converter circuit failure diagnosis based on marginoptimized Random Forests.Removing the Decision tree which has bigger error in the forest based on margin of Decision tree,and then a Random Forest classifier consisting of 200 weighted decision trees is formed.Put the test data into the Random Forest classifier.The result is the fault categories of most votes.The results show that the improved Random Forest classification equipment has strong generalization ability,fast diagnosis rate and high diagnosis precision characteristics.(4)Building the experimental platform which is available to actual test and verify thevarious types of three phase six pulse rectifier circuit fault.Simulate the various types of three phase six pulse rectifier circuit fault in the experimental platform.Put the fault data into BP-Adaboost strong classifier and the improved classifier of random forest,and then verify the diagnostic performance of the two classifiers.