Abstract: Addressing the challenges faced by existing state of charge (SOC) estimation methods for energy storage batteries, such as insufficient accuracy, unclear underlying mechanisms, and poor discriminability in voltage plateau regions, this study focuses on the 330 Ah large-capacity LFP battery. By integrating equivalent models with data-driven methods, we propose an EIS-based SOC estimation approach suitable for early-stage battery aging. First, EIS data were collected at different SOC levels. By comparing the fitting performance of multiple equivalent circuit models, the R₀+(R_p+CPE)||C model was selected as the optimal model for the 330 Ah large-capacity battery. Key impedance characteristics were accurately identified, and their evolution patterns over the 0–100% SOC range were clarified. Given the high sensitivity and strong correlation of the CPE-T parameter with SOC, a fast SOC estimation model was constructed by combining the XGBoost algorithm and polynomial fitting. The results show that for SOC estimation over 0–750 cycles, the model achieved RMSE values ranging from 3.85% to 6.67% and an MAE not exceeding 6%; it also performed stably across different individual cells, meeting the engineering application requirements of BMS for SOC estimation accuracy during early-stage battery aging.