Abstract: Precise estimation of lithium battery state of charge (SOC) has a profound impact on the safe and stable driving of pure electric vehicles. The estimation of lithium battery SOC state mainly includes two hot issues: parameter identification algorithm and SOC estimation algorithm. Aiming at the "data saturation" phenomenon in the identification process and the filtering divergence problem in lithium battery SOC state estimation, this paper proposes a joint algorithm of adaptive forgetting factor recursive least squares (ARWLS) and adaptive unscented kalman filter (AUKF). First, the mathematical model of the second order R-C lithium battery is established. Aiming at the "data saturation" phenomenon in the parameter identification process of the traditional least squares method, an adaptive forgetting factor is introduced to dynamically modify the weight of the new and old data, so as to improve the accuracy and efficiency of online parameter identification. Secondly, aiming at the filtering failure problem of unscented kalman filter, an adaptive unscented kalman filter algorithm is proposed to adaptively system noise and observation noise, so as to improve the adaptability and robustness of SOC estimation. Finally, three SOC estimation algorithms, extended kalman filter (EKF), unscented kalman filter (UKF) and AUKF, are simulated and compared under hybrid pulse power characterization (HPPC) working condition. The simulation results show that the SOC curve estimated by AUKF algorithm has the best performance following the change of the true value curve of SOC, and the estimation accuracy is also better than the other two algorithms, with smaller estimation error and the best convergence.

Key words: Lithium battery; State of charge; On line parameter identification; ARWLS; AUKF

摘要： 精确估计锂电池荷电状态（SOC）对纯电动汽车的安全稳定行驶有着深远影响，对锂 电池 SOC 状态的估计主要有参数辨识算法和 SOC 估计算法两个热点问题。针对辨识过程中 出现的“数据饱和”现象以及锂电池 SOC 状态估计时的滤波发散问题，文章提出了自适应遗 忘因子递推最小二乘法（ARWLS）-自适应无迹卡尔曼滤波（AUKF）联合算法。首先建立了 二阶 R-C 锂电池数学模型，并针对传统最小二乘法在参数辨识过程中出现的“数据饱和”现 象，引入了自适应遗忘因子动态修正新旧数据权重，提升在线参数辨识的准确度以及效率。 其次，针对无迹卡尔曼滤波存在的滤波失效问题，提出了自适应无迹卡尔曼滤波算法来自适 应调整系统噪声和观测噪声，从而提高 SOC 估计时的适应性和鲁棒性。最后在混合动力脉冲 能力特性（HPPC）工况下对扩展卡尔曼滤波（EKF）、无迹卡尔曼滤波（UKF）和 AUKF 三 种 SOC 估计算法进行仿真比较，仿真结果表明，AUKF 算法估计的 SOC 曲线跟随 SOC 真实 值曲线变化的性能最好，估计精度也优于其他两种算法，具有更小的估计误差，收敛性也最 好。

关键词: 锂电池；荷电状态；在线参数辨识；ARWLS；AUKF