Abstract: As one of the core functions of a battery management system (BMS), state-of-charge (SOC) estimation for lithium-ion batteries can effectively prevent overcharging and overdischarging, thereby extending battery service life. Considering the respective advantages and limitations of equivalent circuit and electrochemical models, this study begins with battery modeling and establishes a new coupling model by deriving the electrochemical mechanism model and integrating it with the equivalent circuit model. After establishing the electrochemical model, the differential equations of this complex model were simplified using Padé approximation, converting the nonlinear equations into a more tractable polynomial form. This approach not only improves model accuracy but also ensures good real-time performance while reflecting the internal reaction mechanisms of the battery. For parameter identification of the coupling model, the Levenberg–Marquardt (LM) nonlinear least-squares method was employed due to its weak dependence on initial value settings. This method was used to identify 22 parameters within the model. Additionally, the Kalman filtering algorithm was improved based on the coupling model. The original equivalent circuit model was replaced with the coupling model, incorporating more electrochemical parameter information to enhance model accuracy. Furthermore, the open-circuit voltage (OCV) of the battery was derived from the relationship between the Li⁺ concentration in the electrochemical model and the open-circuit voltages (OCV_i)of the positive and negative electrodes. With accurately identified coupling model parameters and a higher-fidelity battery model, the OCV derived from experimental data was replaced with the model-based OCV, enhancing the Kalman filter algorithm’s accuracy. This replacement also mitigated the impact of sampling errors and hysteresis. After the simulation of UDDS (Urban dynamometer driving schedule), FUDS (Federal urban driving schedule), and DST (Dynamic steering test) conditions, the average absolute error was only 18.6, 28.4, and 24.7 mV, respectively. Based on these simulations, a battery discharge experiment was conducted using a cylindrical lithium-ion battery with a ternary lithium (NCM) positive electrode and a calibrated capacity of 2.5 A·h. A dynamic steering test (DST) current profile was applied, with each 32-min cycle discharging approximately 0.5 A·h (20% SOC), ending after the fifth cycle. The model parameters identified using the LM method were input into the model, and comparisons were made using the traditional FFRLS(Forgotten factor recursive least squares) algorithm with the equivalent circuit model. Simultaneously, SOC estimation was performed using the coupling model and the improved Kalman filter algorithm, and the estimated SOC values were compared with experimental results. Under DST conditions, the extended Kalman filter (EKF) algorithm showed the greatest improvement: the average estimation error was reduced by 1%, significantly enhancing SOC estimation accuracy. The result demonstrates that the coupling and electrochemical models developed in this study preserve the battery’s electrochemical characteristics. Despite incorporating the electrochemical mechanism, the proposed SOC estimation method does not significantly increase runtime, offers strong real-time performance, and enables effective online SOC estimation for lithium-ion batteries.