According to the electrochemical impedance spectroscopy of LiFePO4 battery, an improved PNGV model of double RC parallel connection was proposed. Model parameters were identified under HPPC experiment. The EKF algorithm can’t estimate the noise in real-time when estimating battery SOC, the Sage-Husa adaptive algorithm was introduced into the EKF algorithm, and the adaptive extended kalman filter was obtained. The new algorithm improved the accuracy of SOC estimation by real-time prediction and correction of noise. The simulation models of battery and SOC estimation were built in Matlab/Simulink, and simulation was carried out under the simulated dynamic conditions. The simulation showed that the accuracy of the PNGV model was better than that of the PNGV model; the adaptive extended kalman filter algorithm had faster convergence speed and higher estimation accuracy when compared with the EKF algorithm. The improved model and algorithm had achieved better results.