The invention discloses an unsupervised model parameter migration rolling bearing life prediction method, and belongs to the technical field of rolling bearing state identification and residual life prediction. The method is provided for solving the problems that in practice, rolling bearing labeled vibration data under a certain working condition is difficult to obtain, health indexes are difficult to construct, and the service life prediction error is large. The method comprises the steps that firstly, extracting root-mean-square features from rolling bearing full-life-cycle vibration data,and introducing a new bottom-to-top time sequence segmentation algorithm to segment a feature sequence into three states composed of a normal period, a degradation period and a recession period; marking the state information of an amplitude sequence of the vibration signal subjected to fast Fourier transform, taking the amplitude sequence as an input of an improved full convolutional neural network, extracting deep features, and constructing a source-domain model and a state recognition model subjected to fine adjustment through training to realize rolling bearing multi-state recognition; andestablishing a rolling bearing life prediction model by using a state probability estimation method. Experiments prove that the method provided by the invention does not need to construct health indexes, can realize rolling bearing state identification and life prediction under different working conditions under an unsupervised condition, and obtains a better effect.