A prediction method based on improved particle filtering is proposed to improve the low accuracy and poor generalizability of the remaining life prediction of lithium-ion batteries. First, a double Gaussian model is taken as a degradation empirical model to fit the capacity degradation process of lithium-ion batteries. Then the initial parameters of the degradation model are set by using a priori knowledge, and the particle filtering method is used to update the parameters. The Gaussian mixture method for particle resampling is proposed to solve the particle degradation problem. This fits the complex nonlinear distribution and long-tailed distribution of particles in the resampling process, and ensures that the probability density distribution status of the prediction results is uniform and concentrated. Finally, experimental validation is carried out on different datasets, and the results show that the improved particle filtering method proposed has high accuracy and strong robustness.