Abstract: For photovoltaic inverter circuit in the photovoltaic system is complex,and the failure time is short,this paper puts forward a kind of fault diagnosis methodbased on the improved variational mode decomposition（VMD） and the convolutional neural network（CNN）,which can effectively solve the problems that the fault feature extraction is difficult,characteristic parameters of singularity is poor,and the low fault diagnosis rate caused by poor characteristic parameters.Firstly,the software SIMULINK is used to establish the soft fault model of photovoltaic inverter,and relevant parameters are collected as samples.Then,VMD is used for variational modal decomposition of the parameters to obtain some components,and wavelet transform is used to extract the wavelet energy of each modal component to obtain the fault characteristic value and reduce the dimension.Finally,the CNN is used for fault diagnosis,and the results are compared with the traditional VMD-CNN neural network and VMD-BP neural network,which verifies the correctness and accuracy of the softfault diagnosis of photovoltaic inverter using the network,and has certain advantages.