The estimation accuracy of state of charge (SOC) based on battery model mainly depends on the accuracy of the model. Under dynamic conditions, the input current of the battery changes drastically, and the traditional identification method has poor convergence, which leads to the reduction of model accuracy. In order to improve the accuracy of battery model under dynamic conditions, the traditional least square method with forgetting factor (FFRLS) is improved. By setting the accuracy threshold and introducing gradient correction method, an improved recursive least square method with forgetting factor (IFFRLS) is proposed. Online parameter identification is carried out by using the improved algorithm, and a secondorder RC equivalent circuit model is established. Compared with other models established by traditional parameter identification, the effectiveness of IFFRLS in improving the accuracy of the model is verified, and the average error of the model is 0003 8 V. Finally, the models established by different identification methods are combined with EKF algorithm to estimate SOC, and their errors are compared. The results show that the highprecision model identified by IFFRLS can effectively improve the estimation accuracy of SOC, and the error is within 151% under DST condition.