Enhanced state-of-charge estimation of LiFePO4 batteries using an augmented physics-based model

Voltage hysteresis is prominent in LiFePO4 cells and increases the difficulty to estimate state-of-charge (SOC) accurately. This paper proposes an efficient SOC estimation scheme for LiFePO4 cells with an augmented electrochemical model integrated with an empirical hysteresis assumption. The hysteresis phenomenon is separated from the cell voltage and modeled using a single-state hysteresis formulation. The other contributions to voltage are calculated via the physical model. The extended model is simplified via the discrete-time realization algorithm and validated to match closely measured voltage with the root-mean-square-error (RMSE) less than 18 mV. Furthermore, the unscented Kalman filter (UKF) is applied to the reduced-order model for lithium-ion concentration and SOC estimations. Experimental data starting at multiple SOC points (95%, 79%, 69%, 58%, 52%, 33%) are used to verify the estimator. The results demonstrate that the scheme enables fast convergence and high accuracy of the predicted states even when the UKF is intentionally initialized with large errors. The SOC errors are maintained less than 1% (RMSEs <= 0.42%), indicating that the estimator performance is sustained across the flat area of cell open-circuit-voltage. Finally, the significance of the hysteresis incorporation and the robustness of the model-based estimator considering measurement noise and parameter uncertainty are investigated.

Automated summary

This paper proposes an efficient SOC estimation scheme for LiFePO4 cells, which separates the hysteresis phenomenon from the cell voltage and models it using a single-state hysteresis formulation. The scheme achieves fast convergence and high accuracy of the predicted states, as demonstrated through experiments with different SOC points and the use of the unscented Kalman filter for estimation.