With the surging penetration of the energy storage market, LiFePO4 (LFP) batteries are playing an increasingly pivotal role in achieving carbon neutrality and transportation electrification. However, the complex minor loop hysteresis in the open-circuit voltage (OCV) and the wide voltage plateau during the middle state-of-charge (SOC) range pose significant challenges to accurate modeling and state estimation. Hysteresis, defined as the difference in SOC-OCV curves during charge and discharge processes, is particularly prominent in LFP batteries. Accurate modeling and state estimation that account for hysteresis are essential for their safe and effective application.
To address these challenges, this study proposes a data-driven hysteresis model utilizing historical SOC information and designs an adaptive extended Kalman filter tailored to accommodate hysteresis voltage slope variations. The hysteresis model, developed using deep long short-term memory neural networks trained on hysteresis test data, captures complex voltage hysteresis across different charge/discharge paths. This OCV component is then integrated into a second-order equivalent circuit model to achieve precise modeling. Subsequently, a multistep parameter identification method incorporating a meta-heuristic algorithm is employed to optimize the model parameters. Based on the fusion model, the SOC estimator updates its covariance matrices according to the voltage slope variations during charging from different SOC levels on the plateau, thereby enhancing Kalman gain matching to mitigate cumulative errors and improve accuracy.
Experiments were conducted under various charge/discharge scenarios and at different temperatures. The effectiveness of the proposed method was validated through comparisons with other state-of-the-art methods.