This study simultaneously considers the state‐of‐charge (SOC) estimation and model parameter identification of lithium‐ion batteries with outliers in measurements. Conventional Kalman‐type filters may degrade performance in this case since they assume Gaussian‐distributed measurement noise. To improve the SOC estimation accuracy under this condition, a robust normal‐gamma (NG)‐based adaptive dual unscented Kalman filter (NG‐ADUKF) is proposed. First, by modeling the joint distribution of the state and auxiliary variables of the measurement noise as the NG distribution, the unscented Kalman filter (UKF) is integrated with the NG filter to deal with the heavy‐tailed measurement noise. Second, the online parameter identification and SOC estimation are realized simultaneously by alternatively using two NG‐based adaptive UKFs. The performance of the proposed algorithm is validated by the New European Driving Cycle and Urban Dynamometer Driving Schedule tests. Experimental results show that the proposed NG‐ADUKF algorithm has more accurate SOC estimations compared with the dual UKF (DUKF) and the variational Bayes‐based adaptive DUKF (VB‐ADUKF) in the case of mistuning and outliers. Moreover, the proposed method is more computationally efficient than VB‐ADUKF.

中文翻译：

---

基于正态伽玛的自适应双无味卡尔曼滤波器，可用于电池参数和带有重尾测量噪声的荷电状态估计

本研究同时考虑了带有离群值的锂离子电池的荷电状态（SOC）估计和模型参数识别。在这种情况下，传统的卡尔曼型滤波器可能会降低性能，因为它们会假设是高斯分布的测量噪声。为了提高这种情况下的SOC估计精度，提出了基于鲁棒的基于正常伽玛（NG）的自适应双无味卡尔曼滤波器（NG-ADUKF）。首先，通过将测量噪声的状态和辅助变量的联合分布建模为NG分布，将无味卡尔曼滤波器（UKF）与NG滤波器集成在一起，以处理重尾测量噪声。其次，通过交替使用两个基于NG的自适应UKF，可以同时实现在线参数识别和SOC估计。新的欧洲驾驶周期和城市测功机驾驶时间表测试验证了该算法的性能。实验结果表明，与双UKF（DUKF）和基于变数贝叶斯自适应DUKF（VB-ADUKF）相比，提出的NG-ADUKF算法在误解和离群值情况下具有更准确的SOC估计。此外，所提出的方法比VB-ADUKF具有更高的计算效率。