The internal resistance of the lead-acid battery increases as its operating time increases, which will result in the decrease in its capacity and a consequent reduction in service life. Therefore, the accurate assessment and prediction of its useful life is benefit for improving the ability of continuous power supply and operational reliability of substation DC power system. The support vector machine of LIBSVM is a machine learning method that follows the principle of structural risk minimization. It has the characteristic of using support vector machine to establish the predict model of the battery useful life without modeling the detailed mathematical model of battery. So based on studying the basic principle of support vector machine, the method that uses the support vector machine of LIBSVM to model the non-linear mapping between useful life and both the terminal voltage and state of health based on the training sample data of these three state variables is proposed. At the same time, the method of designing the optimal parameters for the regression machine of LIBSVM based on cross-validation is discussed. The experimental results show that the LIBSVM-based lead-acid battery life prediction model has high prediction accuracy. The feasibility of the proposed prediction method is verified as well.