In order to solve the problem of low accuracy of existing on-line SOH estimation methods for lithium batteries, In this paper, we propose an improved particle swarm optimization (IPSO) method to optimize the support vector regression (SVR) model. Firstly, the association analysis of the extracted health indicators is carried out, and the heuristic algorithm is used to optimize the hyper parameters of SVR. At the same time, the weights of the extracted multiple health indicators are optimized, and the SVR model with the optimal parameters trained by the optimal data sequence is obtained. Using three battery data sets B5, B6, B7 of the PCoE Research Center of the National Aeronautics and Space Administration to test and analyze the proposed method, the results show that the average absolute percentage error (MAPE) and root mean square error (RMSE) of the three groups of batteries are reduced by 62.3% and 65.5% respectively after the algorithm is optimized. Finally, by comparing with the existing prediction models, it is proved that the proposed method has higher online estimation accuracy.