Battery Management System (BMS)

LFP State-of-Charge Estimation: Still a Problem Challenges for state-of-charge estimation: Voltage-based method suffers from path-dependent behavior + strong hysteresis + flat OCV plateau Current-based method suffers from high sensitivity to initial SOC + sensor offset/noise + accumulated error. Che et al. keep current-based method as the backbone and periodically reset SOC using short 30600 s) rest segments in normal driving; A ML model uses recent current, voltage, and temperature history to correct the initial SOC

keep error 3.25% over 430 conditions using only low-frequency signals, suitable for on-board BMS.

Source: 1 Che et al., ACS Energy Lett., 2025; 2 Yan et al., eTransportation, 2025.

Yan et al. propose a voltage-based hybrid model: an LSTM hysteresis OCV model SOC history + temperature) + a 2RC ECM, with a slope-adaptive Kalman filter that adjusts covariances by OCV slope.

Across varied temperatures/currents/partial cycles, 95% of samples reach SOC MAE 0.56%;

RMSE is 46% lower than two benchmark state-of-the-art models.

Extracted from the PDF; charts and images may read imperfectly.