As a key part of system energy supply, the end of life of lithium-ion battery often leads to the degradation or failure of the electrical equipment, or even the collapse of the whole system. Therefore, it is increasingly important to study the remaining useful life (RUL) of the battery and predict the failure time in advance. Aiming at the problems of long training time, difficult parameter determination and unstable output results during the life prediction process of lithium-ion batteries, this paper puts forward the Relevance Vector Machine (RVM) which is more suitable for on-line detection and has better generalization ability, sparser parameter and shorter test time, and optimizes relevance vector machine (RVM) through quantum particle swarm optimization (QPSO) to ensure the stability of predicted output results. The results show that the prediction accuracy of the improved relevance vector machine (QPSO-RVM) is up to 99%, the mean absolute error of battery life prediction is about 2% and the root mean square error is about 3%, which verifies the improved algorithm feasibility and superiority.