Traditional similarity life prediction methods ignore the local evolution characteristics of degradation process， which leads to low prediction accuracy. Traditional characteristic indexes in time and frequency domain are difficult to realize early fault monitoring， and local fluctuation is large in the later stage of degradation. The trend fitting strategy of Gaussian function is introduced and an improved similarity matching optimization method is proposed. A Jensen-Renyi divergence health index based on Gaussian mixture model is proposed to accurately track the evolution trend of rolling bearing degradation. Since it is difficult to obtain a large number of degradation signals in real life cycle， a double exponential function model is constructed to simulate degradation signals and verify the validity of the simulation data to expand the reference dictionary set. Gaussian function is used to fit the degradation data and parameter similarity principle is proposed to predict the remaining service life. The experimental results of rolling bearing life cycle degradation verify that the proposed method can effectively improve the prediction accuracy of residual life.