• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2021, Vol. 57 ›› Issue (14): 118-128.doi: 10.3901/JME.2021.14.118

• 特邀专栏:电源系统设计、管理与大数据 • 上一篇    下一篇

扫码分享

基于改进最小二乘支持向量机与Box-Cox变换的锂离子电池容量预测

舒星1, 刘永刚2, 申江卫1, 陈峥1   

  1. 1. 昆明理工大学交通工程学院 昆明 650500;
    2. 重庆大学机械传动国家重点实验室 重庆 400044
  • 收稿日期:2020-06-19 修回日期:2021-03-20 出版日期:2021-09-15 发布日期:2021-09-15
  • 通讯作者: 陈峥(通信作者),男,1982年出生,博士,教授,博士研究生导师。主要研究方向为动力电池管理、智能车辆控制及混合动力汽车能量管理。E-mail:chen@kust.edu.cn
  • 作者简介:舒星,男,1992年出生,博士研究生。主要研究方向为动力电池管理。E-mail:shuxing92@kust.edu.cn
  • 基金资助:
    国家自然科学基金(61763021,51775063)和国家重点研发计划(2018YFB0104000)资助项目

Capacity Prediction for Lithium-ion Batteries Based on Improved Least Squares Support Vector Machine and Box-Cox Transformation

SHU Xing1, LIU Yonggang2, SHEN Jiangwei1, CHEN Zheng1   

  1. 1. Faculty of Transportation Engineering, Kunming University of Science and Technology, Kunming 650500;
    2. State Key Laboratory of Mechanical Transmissions & School of Automotive Engineering, Chongqing University, Chongqing 400044
  • Received:2020-06-19 Revised:2021-03-20 Online:2021-09-15 Published:2021-09-15

摘要: 精确、可靠的电池容量预测可以避免电池滥用,提升电池使用安全;同时在此基础上开展的剩余寿命估测能够为电池系统维护及更换提供参考。基于改进双最小二乘支持向量机方法和Box-Cox变换,提出一种锂离子电池容量及剩余循环寿命的协同估算方法。首先提取老化电池部分容量增量曲线包络面积作为特征量,通过Box-Cox变换进一步提高特征量与目标估计量之间的相关性。然后基于瑞利熵理论改进传统最小二乘支持向量机算法的稀疏性,建立电池容量和剩余使用寿命协同估算模型,结合层次分析法和熵权法对估算结果进行充分地评估。最后,采用粒子群优化算法搜索改进最小二乘支持向量机算法中的最优超参数组合。估算结果显示所研究的方法能够显著提高特征参数与估计量之间的线性相关性,容量估计误差小于1.44%,剩余使用寿命预测误差小于47次循环,验证了算法的有效性。

关键词: 锂离子电池, 电池容量, 剩余使用寿命, 最小二乘支持向量机, Box-Cox变换

Abstract: Accurate and reliable estimation of battery capacity can help prevent overuse and ensure operation safety; and on this basis, the remaining useful life (RUL) prediction can supply the guidance for maintenance and replacement battery systems. A synchronous estimation method for capacity and RUL is proposed based on the improved dual least squares support vector machine (LS-SVM) and Box-Cox transformation. First, the aging feature variables are extracted from the envelope area of partial incremental capacity curves, and the Box-Cox transformation is employed to improve the correlation between the aging features and target estimation variables. Then, the Renyi entropy is applied to improve the sparseness of traditional LS-SVM; and a joint estimation model for capacity and RUL is constructed. The estimated performance is sufficiently evaluated by applying the analytic hierarchy process and the entropy weight methods. Finally, the particle swarm optimization (PSO) is exploited to search the optimal hyper-parameter combination of least squares support vector machine. The experimental results demonstrate that the proposed method can significantly improve aging feature performance, and the estimation error of capacity and RUL can be respectively restricted within 1.44% of nominal capacity and 47 cycles, thereby verifying the effectiveness of the proposed method.

Key words: lithium-ion batteries, battery capacity, remaining useful life (RUL), least squares support vector machine (LS-SVM), Box-Cox transformation

中图分类号: