• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Charging amp; infrastructure symposium 2018 >High Accurate Lithium-Ion Battery SOC Estimation with Data-Driven Battery Model
【24h】

High Accurate Lithium-Ion Battery SOC Estimation with Data-Driven Battery Model

机译：数据驱动电池模型的高精度锂离子电池SOC估计

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Lithium-ion (Li-ion) battery technology has been developed fast during the past few decades, which facilities the application of Li-ion battery in various energy storage systems. The efficiency and safety of the battery pack are closely related to the battery State of Charge (SOC). Our contributions are mainly focusing on the battery modeling and SOC estimation, and we summary them into three aspects: 1) . The industry is more focusing on online implementable SOC estimation methods. Depending on their governing principles, the online SOC estimation algorithms are divided into five categories by us: Coulomb Counting Methods (CCMs); Open Circuit Voltage Methods (OCVMs); Impedance Spectroscopy Based Methods (ISBMs); Model Based Methods (MBMs) and Artificial Neural Network Based Methods (ANNBMs). CCMs and OCVMs are traditional SOC estimation method, which is more simple and effective in SOC calculation. However, CCMs need accurate current measurement and OCVMs require long battery relaxation time. Although ISBMs is a nondestructive method, the exact relationship of EIS and SOC and the repeatability of EIS for online measurement still need further research. MBMs rely on precise battery model for SOC estimation. It is difficult to build an accurate enough model to describe all the battery external characteristics. If appropriate samples are selected and optimized parameters are chosen in the training process, ANNBMs are able to present accurate SOC estimation. Among all the SOC estimation methods, MBMs seem to be the most practical choice for online SOC estimation at present. For more details, please refer to our publications [1], [2]. 2) . As battery model is important for MBMs, we proposed data-driven models for Li-ion battery. The main idea in this part is to dynamic linearize the battery model by machine learning methods. Combined with Adaptive Unscented Kalman Filter (AUKF) and Support Vector Machine (SVM) in [3], [4], less than 2% average absolute error is obtained for SOC estimation. AUKF is able to utilize the SVM battery model without the necessity of calculating the Jacobin matrix as in Extended Kalman Filter (EKF). Moreover, the proposed SVM model can be updated when enough samples have been collected. In order to further simplify the modeling process, battery models based on Multivariate Adaptive Regression Splines (MARS) and Partial Least Squares (PLS) are also proposed by us in [5] and [6]. 3). Because of the good performance, nonlinear filters (such as, EKF, Unscented Kalman Filter (UKF), Central Difference Kalman Filter (CDKF), Square Root Unscented Kalman Filter (SR-UKF), Square Root Central Difference Kalman Filter (SR-CDKF), Particle Filter (PF), H-infinity filter, etc) are widely investigated in the literature on the basis of a model based structure. The previous mentioned seven nonlinear filters are compared in terms of accuracy and execution time in our work [2]. The absolute error proves that SR-UKF, SR-CDKF, PF and H-infinity filter are more suitable for the strongly nonlinear system. The experimental results have shown that H-infinity filter gives a good compromise in terms of accuracy and execution time for SOC estimation.
机译：在过去的几十年中，锂离子（Li-ion）电池技术得到了快速发展，这为锂离子电池在各种储能系统中的应用提供了便利。电池组的效率和安全性与电池充电状态（SOC）密切相关。我们的贡献主要集中在电池建模和SOC估算上，我们将其概括为三个方面：1）。业界更加关注在线可实施SOC估算方法。根据它们的管理原理，在线SOC估计算法被我们分为五类：库仑计数法（CCM）；开路电压方法（OCVM）;基于阻抗谱的方法（ISBM）；基于模型的方法（MBM）和基于人工神经网络的方法（ANNBM）。 CCM和OCVM是传统的SOC估计方法，在SOC计算中更加简单有效。但是，CCM需要精确的电流测量，而OCVM需要较长的电池松弛时间。尽管ISBM是一种非破坏性方法，但EIS和SOC的确切关系以及EIS在线测量的可重复性仍需要进一步研究。 MBM依靠精确的电池模型来进行SOC估算。很难建立足够准确的模型来描述所有电池的外部特性。如果在训练过程中选择了适当的样本并选择了优化的参数，则ANNBM能够提供准确的SOC估计。在所有SOC估计方法中，MBM似乎是当前在线SOC估计的最实用选择。有关更多详细信息，请参阅我们的出版物[1]，[2]。 2）。由于电池模型对于MBM至关重要，因此我们提出了锂离子电池的数据驱动模型。本部分的主要思想是通过机器学习方法动态线性化电池模型。结合[3]，[4]中的自适应无味卡尔曼滤波器（AUKF）和支持向量机（SVM），获得的SOC估计平均平均误差小于2％。 AUKF能够利用SVM电池模型，而无需像扩展卡尔曼滤波器（EKF）中那样计算雅可宾矩阵。此外，当已经收集了足够的样本时，可以更新所提出的SVM模型。为了进一步简化建模过程，我们在[5]和[6]中还提出了基于多元自适应回归样条（MARS）和偏最小二乘（PLS）的电池模型。 3）。由于性能良好，非线性滤波器（例如EKF，无味卡尔曼滤波器（UKF），中心差卡尔曼滤波器（CDKF），平方根无味卡尔曼滤波器（SR-UKF），平方根中心差卡尔曼滤波器（SR-CDKF） ），粒子滤波器（PF），H-无穷大滤波器等）在基于模型的结构的基础上进行了广泛的文献研究。我们在工作中比较了前面提到的七个非线性滤波器的准确性和执行时间[2]。绝对误差证明SR-UKF，SR-CDKF，PF和H-∞滤波器更适合于强非线性系统。实验结果表明，H无限滤波器在SOC估计的准确性和执行时间方面给出了很好的折衷方案。

著录项

  • 来源
  • 会议地点 Mainz(DE)
  • 作者

    Xiao Cai; Luo Guangzhao;

  • 作者单位

    Xian Stropower Technologies Co. Ltd., Chuanghui Road, Gaoxin District, Xi'an Shaanxi, 710072 China;

    Northwestern Polytechnical University, 127 West Youyi Road, Beilin District, Xi'an Shaanxi, 710072 China;

  • 会议组织
  • 原文格式 PDF
  • 正文语种 eng
  • 中图分类
  • 关键词

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号