Abstract: Lithium-ion battery is becoming the development trend of airborne battery because of its superior performance, and have been applied in civil aircraft models such as B787.The performance of lithium-ion battery deteriorates with service time increasing. When its performance cannot meet the airworthiness requirements, the life of lithium-ion battery will end, which will affect the performance and safety of aircraft electrical system.Therefore, accurately predicting the remaining useful life (RUL) of lithium-ion batteries is of great significance for timely maintain or replace batteries and ensure flight safety.Aiming at RUL prediction for lithium-ion battery, this article extracts features with incremental capacity analysis and trains RUL prediction models based on the improved grey wolf optimization (IGWO) and support vector regression (SVR).Furthermore, IGWO algorithm is proposed for solving the problem that grey wolf optimization （GWO） is prone to stagnation at local optima.In the optimization process of GWO, the ordinary gray wolves only search for new optimal solution under the guidance of three leader wolves. IGWO adds memory ability to each gray wolf by rewriting the position update equation, so that each gray wolf's historical optimal solution will play a role during optimization process,which enhances the algorithm's global search and convergence capabilities. Furthermore, IGWO combines Levy flight into the position update equation and proposes a new random dynamic control factor, which makes the search scope of gray wolves expand and is beneficial to avoiding local optimization.IGWO generates chaotic sequences based on Skew Tent mapping to optimize the initial position distribution of wolves.In order to compare the optimization ability of GWO before and after the improvement, this paper carries out a optimization comparison experiment based on 19 commonly used benchmark functions. The results show that IGWO algorithm can effectively avoid getting stuck at locally optimal value, with faster convergence speed and better optimization effect compared to GWO for almost all functions. Based on NASA lithium-ion battery dataset, the RUL prediction and performance assessment abilities of IGWO-SVR, GWO-SVR and SVR are compared. The results show that the model trained with IGWO-SVR achieves higher prediction accuracy on the data of all four batteries, and the root mean square error of the prediction results is reduced by more than 10% compared with GWO-SVR.