To address the low accuracy in diagnosing faults in wind turbine bearings caused by the different characteristic distribu? tion of the source domain data and the target domain data， a fault transfer diagnosis method using improved residual neural net? works is proposed. The convolution kernel and pooling kernel are set to a size suitable for the convolution operation of one-dimen? sional signals， allowing for direct extraction of fault features from the bearing vibration signals； Both batch normalization and case normalization are used in the one-dimensional residual network to further enhance the feature extraction ability of the model； In the model training stage， a new loss function is constructed based on the multiple kernel maximum mean discrepancy between the source domain data and the target domain data to improve the transfer learning and classification ability of the model. The effective? ness of the method is verified by conducting the experimental data of the faulty bearings. The results show that the proposed meth? od can effectively extract the important features of bearing faults and achieve the transfer diagnosis and accurate classification of the bearing faults. This holds true even under varying speed operation conditions and when the bearing fault vibration signals are dis? turbed by some noise components. Therefore， this work provides a useful strategy in developing intelligent fault diagnosis technolo? gy of rotating machinery under complex working conditions.