Abstract: The lithium-ion batteries have drawn much attention as the major energy storage system. However, the battery state estimation still suffers from inaccuracy under dynamic operational conditions, with the unstable environmental noise influencing the robustness of estimation. This paper presents a 5th-order cubature Kalman filter with improvements on adaptivity for real-time state-of-charge estimation. The second-order equivalent circuit model is developed for describing the characteristics of battery, and parameter identification is carried out according to particle swarm optimization. The developed method is validated in stable and dynamic conditions, and simulation results show a satisfactory consistency with the experimental results. The maximum estimation error under static conditions is less than 3% and the maximum error under dynamic conditions is 5%. Numerical analysis indicates that the method has better convergence and robustness than the traditional method under the disturbances of initial error, which demonstrates the potential for EV applications in harsh environments. The proposed method shows application potential for both online estimations and cloud-computing system, indicating its diverse application prospect in electric vehicles.