Addressing the OCV-SOC Flat Region in LFP Cells for SOC Estimations Using EIS Measurements

Lithium Iron Phosphate (LFP) batteries are characterized by a flat voltage profile across a wide state-of-charge (SOC) range, which reduces the effectiveness of open-circuit voltage (OCV) curves for SOC estimation, particularly in the mid-SOC region. To address this limitation, comprehensive Electrochemical Impedance Spectroscopy (EIS) experiments were performed on LFP cells across the full SOC range and under multiple temperature conditions. From the experimental dataset, key impedance features were identified using a Random Forest algorithm, and a convolutional neural network with elastic net regularization (CNN-EN) was developed to model SOC. The CNN-EN model achieved high predictive accuracy, with an MSE of 0.0012, MAE of 0.02753, R 2 of 0.9820, and MAPE of 0.0607. Notably, a single frequency point at 0.0125 Hz delivered performance comparable to the full EIS spectra, reducing data collection time by approximately 85% and model training time by 6% while maintaining accuracy. These results demonstrate that experimentally validated EIS features, when combined with machine learning, provide a robust, efficient, and accurate pathway for SOC estimation in LFP batteries, with strong potential for integration into real-time battery management systems (BMS).