Abstract: As one of the core functions of BMS, the State of Charge (SOC) estimation of lithium-ion battery can effectively avoid the problem of overcharge and overdischarge of the battery, thus extending the battery life. According to the advantages and disadvantages of the equivalent circuit model and the electrochemical model, a coupling model is established in this paper, which can improve the accuracy of the model, ensure good real-time performance, and reflect the internal reaction mechanism of the battery in real time. Based on the coupling model, the LM nonlinear least square method is used to identify 22 parameters in the model. Secondly, the Kalman filter algorithm is improved based on the coupling model, and the model parameters and the open circuit voltage curve calculated by the electrochemical model are replaced by the experimental values, which avoids the influence of sampling error and hysteresis characteristics. In the simulation of UDDS, FUDS and DST conditions, the average absolute error is only 18.6mV, 28.4mV and 24.7mV. On this basis, the paper designed a battery discharge experiment. Under the experimental DST current condition, the EKF algorithm improved the most, the average error was reduced by 1%, and the SOC estimation error was effectively improved. The experimental results show that although the electrochemical mechanism is added, the estimated running time is not increased too much, which has good real-time performance, and the lithium battery SOC can be well estimated online in real time.