Aiming at the problem that a single time-frequency domain index cannot fully interpret the degradation characteristics of rolling bearing life cycle and it is difficult to predict the remaining useful life（RUL）， a RUL prediction method is proposed combining mean square harmonic noise ratio （MSHNR） and regularized particle filter （RPF）. Firstly， the feature extraction of the whole bearing degradation process is realized based on MSHNR. Then， the state space models of rolling bearings at different degradation stages are established based on Paris model and Foreman model. Aiming at the problem that particle filter is easy to exhaust particles and weaken diversity during resamping，the resamping process of regularized particle filter （RPF） is improved by using Euclide-based kernel function， so as to track the degradation state. Finally， the effectiveness of the proposed method is verified by the accelerated life test data set of rolling bearings. Relevant research results can provide reference for predictive maintenance of rolling bearings in nuclear power rotating equipment， and improve the public's understanding and trust in the operation of nuclear power rotating equipment.