Research On Fault Diagnosis Algorithm Of Fusion Power Supply

Fusion device is a complex system,and the fusion power supply is an important support system for the stable operation of the device.As a key component of the fusion pulse power supply,the converter will affect the normal operation of the power supply.Therefore,accurate and fast fault diagnosis is very important for the healthy operation of the power converter.With the rapid development of computer technology,the intelligent fault diagnosis with the help of deep learning theory has gradually become an important research direction.Firstly,this paper analyzes the advantages of deep neural network in electrical system fault diagnosis,and expounds the characteristics of convolutional neural network(CNN).Based on the theory of magnetic compression,the electrical parameters of EAST discharge#41195 are analyzed,which provide a reference design for the high-performance plasma vertical field,and lead to the demand of Magnetic Compression Coil Power Supply(MCCPS).The fault diagnosis of the power converter is critical to achieving proper operation and good control of the power supply.The fault diagnosis method of the power converter based on deep convolutional neural network requires a large amount of fault data.The Simulink model is also built with reference to the classic fusion power converter model,the corresponding short-circuit fault is inserted to obtain the fault data.In addition,the data enhancement is realized by the Overlap method.The 1D-CNN model for classification of working conditions is proposed.The operating conditions of the magnetic field power converter are classified,and the accuracy rate is 98%.Taking the circulation condition as the research object,the 1D-CNN model suitable for fault diagnosis of power converters is proposed,which can automatically complete feature extraction and fault identification.Using the data visualization technology,the process of diagnosing short-circuit fault signals by the 1D-CNN model is demonstrated.According to the characteristics of the short-circuit fault signal,the 1D-BNCNN model is further proposed due to the slow training problem of the 1D-CNN model.The Batch Normalization layer is added to the model,so that the statistical distribution of the input data of each convolution layer is relatively close,which is beneficial to the training speed and accuracy of the network.The results show that after 20 iterations,the model converges and the fault recognition rate reaches 95%.Aiming at the deficiency of the statistical data for the measured data of the fusion vertical magnetic field power supply,a transfer learning model based on 1D-BNCNN is presented.The model follows the 1D-BNCNN model structure,and small sample test data is adopted to realize the model transferring.The transferring model is used for the fault diagnosis of the coil power converter simulation of the LHD device.The accuracy of the model after transferring is 91%.This application demonstrates the generalization potential of such model for fault diagnosis of fusion power converters.