In response to the weak generalization of cross-domain fault diagnosis for rolling bearings in wind turbine units, a Wind Turbine Bearing Fault Diagnosis Method via Combined Features Adaptation is proposed. Firstly, a data selection method based on one-dimensional convolutional neural networks is introduced to filter data from the source domain that is similar to the target domain, forming a new source domain to reconstruct the training dataset and mitigate the negative transfer effect. Secondly, the continuous wavelet transform is applied to obtain two-dimension time-frequency features of the training data, these features are afforded for the residual neural network, which generates one-dimensional feature vectors as the initial features for transfer learning, ensuring comprehensive extraction of the training data's features. Thirdly, a new combined feature adaptation loss function is presented. In it, not only the feature transfer loss of the source domain and the target domain, but also the classification loss of the source domain is considered. This enables effective feature transfer from the source domain to the target domain. Finally, the softmax classifier of the constructed residual neural network is used for the diagnosis classification of the fault. By the simulation under different kinds of migration tasks between source domain and target domain, the proposed model is of higher diagnosis performance in the cross-domain environment.