To solve the problem of low prediction accuracy of remain useful life (RUL) of rolling bearings caused by limited information of single source domain and insufficient alignment granularity of domain adaptation, Multi-source and subdomain adaption network (MS_SAN) prediction method for residual life of rolling bearings was proposed. This paper has said, the collected original vibration signal is processed by FFT, and the frequency domain signal is input into the model. Perform one-dimensional mapping to visualize data of multiple target domains and source domains in the common feature space. Then adjust the degradation stage of each target domain and source domain to meet the independent requirements of the feature space. Finally, combined with the output of RUL prediction module in various fields, the final prediction results are obtained. The experimental results on the real world dataset PHM2012 show that the proposed model MS_ The performance of SAN is superior to the most advanced methods, and it can effectively predict its remaining use value.