Most methods for predicting the Remaining useful life of deep learning of rotating units usually assume that the data distribution of training data and test data is the same, resulting in low prediction accuracy of the model under different working conditions. For the above problems, this paper proposed a model transfer method based on adversarial training, where the transfer object is a rotating multi-unit RUL prediction model. Aiming at the transfer scenario where the source domain and target domain have different working conditions and the target domain lacks label samples, a domain classifier was introduced to extract the common features of the source domain and target domain data. The feature extraction network in the RUL prediction model was retrained by combining the labeled data in the source domain with the unlabeled data in the target domain. In the training process, auto association and correspondence constraints were added to improve the ability to extract common features, thus realizing the migration application of the model in different scenarios. The test results of the transfer model using the XJTU-SY public dataset revealed that the prediction accuracy of the method described in this paper is higher than that of the original prediction model under the new working conditions. Compared with other transfer methods, the prediction error of this method is smaller, and it has a better effect on predicting the remaining useful life of rotating units under variable working conditions.