To address the problem of inaccurate parameters in equivalent circuit models of lithium batteries and uncertainty in complex operating noise leading to the low accuracy in state-of-charge (SOC) estimation. An adaptive extended Kalman filtering (AEKF) method incorporating recursive least squares with forgetting factor (FFRLS) is proposed. In each step of the SOC estimation process, the FFRLS algorithm is first used to identify the parameters of the first-order RC equivalent circuit model in real time following the changes of the experimental working environment, which increases the model accuracy and accurately describes the dynamic characteristics of the Li-ion battery when it is in operation. Then the AEKF algorithm is used to update and correct the system noise in real time and estimate the SOC on-line. The experimental platform for power lithium batteries was designed and constructed. Under both the dynamic stress test (DST) and the beijing bus dynamic stress test (BBDST) operating conditions, the maximum absolute error of the method's estimation is lower than 0.15%, the average absolute error is below 0.077, and the root mean square error is below 0.007. Compared with the extended kalman filtering (EKF) method, the estimation effect of the proposed method is greatly improved.