储能科学与技术 ›› 2023, Vol. 12 ›› Issue (1): 209-217.doi: 10.19799/j.cnki.2095-4239.2022.0508

• 储能测试与评价 • 上一篇    下一篇

模型与数据双驱动的锂电池状态精准估计

陈清炀1(), 何映晖1, 余官定1(), 刘铭扬2, 徐翀2, 李振明2   

  1. 1.浙江大学信息与电子工程学院，浙江 杭州 310058
    2.中国电力科学研究院有限公司储能 与电工新技术研究所，北京 100192
  • 收稿日期:2022-09-07 修回日期:2022-09-23 出版日期:2023-01-05 发布日期:2023-02-08
  • 通讯作者: 余官定 E-mail:22231156@zju.edu.cn;yuguanding@zju.edu.cn
  • 作者简介:陈清炀（2000—），女，硕士研究生，主要研究方向为锂电池热失控预警，E-mail：22231156@zju.edu.cn；
  • 基金资助:
    国家电网有限公司“储能锂离子电池智能监测技术研究”科技项目(5500-202255364A-2-0-ZN)

Integrating model- and data-driven methods for accurate state estimation of lithium-ion batteries

Qingyang CHEN1(), Yinghui HE1, Guanding YU1(), Mingyang LIU2, Chong XU2, Zhenming LI2   

  1. 1.College of Information and Electronic Engineering, Zhejiang University, Hangzhou 310058, Zhejiang, China
    2.Energy Storage and Novel Technology of Electrical Engineering Department, China Electric Power Research Institute Co. , Ltd. , Beijing 100192, China
  • Received:2022-09-07 Revised:2022-09-23 Online:2023-01-05 Published:2023-02-08
  • Contact: Guanding YU E-mail:22231156@zju.edu.cn;yuguanding@zju.edu.cn

RichHTML

37

PDF

266

摘要：

针对电池荷电状态估计常用的模型驱动法与数据驱动法的缺点，本工作提出了一种模型与数据双驱动的锂电池状态精准估计算法。在建立经典二阶电池模型后，先使用扩展卡尔曼滤波器与无迹卡尔曼滤波器组成的双卡尔曼滤波器进行初步的锂电池系统状态估测，再将初步的估算结果输入LSTM神经网络实现误差纠正，得到最终估测结果。本工作利用来自NASA PCoE的电池数据集对单驱动算法和双驱动算法分别进行了性能测试，结果表明双驱动法在降低了估算系统对数据依赖性的同时提高了估算精度以及算法鲁棒性，结合了两种单驱动法的优点并弥补了各自的缺点，得到了较为优异的结果。

关键词: 锂电池, 电池荷电状态, 电池健康状态, 模型驱动法, 数据驱动法, 扩展卡尔曼滤波, 无迹卡尔曼滤波, LSTM神经网络

Abstract:

Addressing the inadequacies of the conventional model- and data-driven methods, an integrating strategy combining both methods, for accurate state estimation of lithium-ion batteries is proposed for estimating battery state-of-charge. After establishing the classical second-order battery model, a dual-Kalman filter, composed of an extended Kalman filter and an unscented Kalman filter, was used to estimate the status of the lithium battery system preliminarily. Then, the preliminary estimation results were input into the LSTM neural network to correct the errors and complete the data-driven part. Datasets from NASA PCoE were used to test the performance of the single-and dual-driven methods. Results show that the integrating method reduces the dependence of the estimation system on the data while improving the estimation accuracy and robustness because it combines the advantages of the model-and data-driven methods and makes up for their shortcomings. Satisfactory results were obtained.

Key words: lithium battery, state of charge, state of health, model-driven method, data-driven method, extended Kalman filter, unscented Kalman filter, long-short-term neural network

中图分类号: