Aiming at the power battery under complex and variable working conditions, offline parameter identification cannot reflect the dynamic characteristics of the battery in real time, resulting in low parameter identification accuracy, and the untraceable Kalman filter (UKF) is very limited to deal with the noise when estimating the state of charge (SOC) of the battery, and at the same time, the non-positivity problem in the processing of the covariance matrix leads to algorithm fluctuations and estimation failures. In this paper, based on the dual polarization (DP) circuit model, the forgotten factor recursive least squares (FFRLS) and singular value decomposition-adaptive untraceable Kalman filter (SVD-AUKF) methods for online estimation of battery SOC are proposed. The simulation results show that the mean absolute error (MAE) and root mean square error (RMSE) of SVD-AUKF for simulation validation are 0.528 6 % and 0.544 7 %, respectively, when comparing with the real SOC values under the complex operating conditions (U.S.Federal Urban Operating Conditions (FUDS)), and the MAE and RMSE are improved on the basis of traditional UKF algorithm, respectively, by 57.96 % and 63.3 %, further indicating that SVD-AUKF is more accurate and stable.