Accurately identifying the aging-related parameters of a lithium-ion electrochemical model is crucial for the advanced battery management systems over the cells’ service life. However, the multiparametric and highly nonlinear mathematical structures of the physical model heighten the difficulty for parameterization. Thus, analyzing the influence of degraded parameters on model output holds the key to efficient identification. In this article, a statistics-based global sensitivity analysis of overall 16 aging parameters in the pseudo-2-D model of lithium-ion batteries is investigated under both the charge process and dynamic driving cycles at 10°, 25°, and 45°. First, massive samples of the parameters are generated synchronously with the Latin hypercubes method. Then the model is simulated with the Monte Carlo technique. Finally, the sensitivity of each parameter is ranked with the partial correlation coefficient which quantifies the relation between the parameter variations and the voltage residuals. The results turn out that the sensitivity of each parameter varies at different operating conditions. Specifically, the resistance-related parameters are the most sensitive than capacity and diffusion-related parameters. To validate the effectiveness of the proposed approach, 16 aging parameters are clustered for identification. The identified model achieves low voltage root-mean-square errors across 170 cycles at three temperatures.

中文翻译：

---

考虑老化的锂离子电池电化学模型全局参数敏感性分析

准确识别锂离子电化学模型的老化相关参数对于先进的电池管理系统在电池的使用寿命内至关重要。然而，物理模型的多参数和高度非线性的数学结构增加了参数化的难度。因此，分析退化参数对模型输出的影响是有效识别的关键。在这篇文章中，在 10°、25° 和 45° 的充电过程和动态驱动循环下，研究了锂离子电池伪二维模型中所有 16 个老化参数的基于统计的全局灵敏度分析. 首先，大量的参数样本与拉丁超立方体方法同步生成。然后用蒙特卡罗技术模拟模型。最后，每个参数的灵敏度与偏相关系数排序，偏相关系数量化了参数变化与电压残差之间的关系。结果表明，每个参数的灵敏度在不同的操作条件下会有所不同。具体而言，阻力相关参数比容量和扩散相关参数最敏感。为了验证所提出方法的有效性，对 16 个老化参数进行了聚类以进行识别。确定的模型在三个温度下的 170 个周期内实现了低电压均方根误差。结果表明，每个参数的灵敏度在不同的操作条件下会有所不同。具体而言，阻力相关参数比容量和扩散相关参数最敏感。为了验证所提出方法的有效性，对 16 个老化参数进行了聚类以进行识别。确定的模型在三个温度下的 170 个周期内实现了低电压均方根误差。结果表明，每个参数的灵敏度在不同的操作条件下会有所不同。具体而言，阻力相关参数比容量和扩散相关参数最敏感。为了验证所提出方法的有效性，对 16 个老化参数进行了聚类以进行识别。确定的模型在三个温度下的 170 个周期内实现了低电压均方根误差。