Abstract: Aiming at the problem that the vibration signals collected by different types of wind turbine generator rolling bearings have different distribution and the sample labels of rolling bearings to be diagnosed are insufficient,this paper proposed a fault diagnosis method of wind turbine generator rolling bearings based on clustering domain adaptive convolutional neural network（CDA-CNN）.Firstly,the features of labeled bearing data in the source domain and unlabeled bearing data in the target domain were extracted by using the 1D convolutional neural network. Secondly,the clustering method was used to reduce the difference in the conditional distribution of data features and provided pseudo labels for target domain data. Then,the maximum mean difference（MMD）was used to align the edge distribution of the two domains. Finally,the fault diagnosis model of wind turbine generator rolling bearings was obtained.The proposed CDA-CNN is applied to the fault diagnosis of actual wind turbine generator rolling bearings. The diagnosis results show that the fault diagnosis accuracy of the proposed method is as high as 92.52%,which effectively solves the problem of insufficient available data labels. The test results show that the diagnostic accuracy and transfer of CDA-CNN are better than other methods,and it has a certain engineering application value for the fault diagnosis of wind turbine generator rolling bearings.