Abstract: To address the issue of limited state of charge (SOC) estimation accuracy for lithium-ion batteries under complex dynamic operating conditions, caused by time-varying model parameters, this paper proposes a joint estimation method based on an improved variable forgetting factor recursive least squares (VFF-RLS) algorithm and unscented Kalman filter (UKF). First, a secondorder resistor-capacitor (RC) equivalent circuit model is established, and the VFF-RLS algorithm is employed to identify the model parameters in real time. Then, an adaptive adjustment of the forgetting factor is achieved by constructing a function based on voltage prediction errors, which resolves the trade-off between convergence speed and steady-state fluctuation inherent in traditional fixed-parameter algorithms. Subsequently, the real-time identified parameters are incorporated into the UKF algorithm for SOC estimation, effectively suppressing nonlinear errors. Finally, experimental validation is conducted. The results demonstrate that under the hybrid pulse power characteristic (HPPC) condition, the root mean square error (RMSE) of the model terminal voltage prediction is only 0.001 3 V, confirming the accuracy of the parameter identification. Under the dynamically severe dynamic stress test (DST) condition, the SOC estimation RMSE of the proposed algorithm is 0.744 6%, showing a significant improvement in accuracy compared to the fixed forgetting factor algorithm. Under the complex Urban dynamometer driving schedule (UDDS) condition, the algorithm maintains a high estimation accuracy of 0.791 9% without the need for manual parameter tuning. Comparative experiments confirm that the proposed joint algorithm outperforms the traditional extended Kalman filter (EKF) algorithm in both accuracy and robustness.

Key words: lithium-ion batteries; SOC estimation; second-order RC model; VFF-RLS; unscented Kalman filter

摘要： 针对锂离子电池在复杂动态工况下因模型参数时变导致荷电状态（SOC）估算精度受 限的问题，文章提出一种基于改进变遗忘因子递推最小二乘法（VFF-RLS）与无迹卡尔曼滤 波（UKF）的联合估算方法。首先，建立二阶电阻-电容（RC）等效电路模型，利用 VFF-RLS 算法实时辨识模型参数。然后，通过构建基于电压预测误差的函数自适应调节遗忘因子，解 决了传统固定参数算法在收敛速度与稳态波动之间的矛盾。进而，将实时参数代入 UKF 算法 进行 SOC 估算，以抑制非线性误差。最后，通过实验进行验证，结果表明，在混合脉冲功率 特性（HPPC）工况下，模型端电压预测均方根误差（RMSE）仅为 0.001 3 V，验证了辨识的 准确性；在动态剧烈的动态应力测试（DST）工况下，算法的 SOC 估算 RMSE 为 0.744 6%， 相比固定遗忘因子算法精度显著提升；在复杂的城市道路循环工况（UDDS）工况下，算法无 需人工调参即可保持 0.791 9%的估算精度。对比实验证实，该联合算法在精度和鲁棒性上均 优于传统的扩展卡尔曼滤波（EKF）算法。

关键词: 锂离子电池；SOC 估算；二阶 RC 模型；VFF-RLS；无迹卡尔曼滤波