储能科学与技术 ›› 2020, Vol. 9 ›› Issue (5): 1566-1573.doi: 10.19799/j.cnki.2095-4239.2020.0022

• 储能测试与评价 • 上一篇    下一篇

基于EEMD-GSGRU的锂电池寿命预测

易灵芝1,3(), 张宗光1,2(), 范朝冬1,2, 罗显光4, 李旺4, 刘文翰1,2   

  1. 1.智能计算与信息处理教育部重点实验室(湘潭大学
    2.湖南省多能源协同控制技术工程研究中心,湖南 湘潭 411105
    3.湖南省风电装备与能源变换2011协同创新中心,湖南 湘潭 411101
    4.大功率交流传动电力机车系统集成国家重点实验室,湖南 株洲 412001
  • 收稿日期:2020-01-08 修回日期:2020-05-25 出版日期:2020-09-05 发布日期:2020-09-08
  • 通讯作者: 易灵芝 E-mail:ylzwyh@xtu.edu.cn;1574511421@qq.com
  • 作者简介:张宗光(1992—),男,硕士研究生,主要研究方向为锂电池寿命预测,E-mail:1574511421@qq.com
  • 基金资助:
    国家自然科学基金项目(61572416);湖南省自然科学基金项目(2020JJ6023);大功率交流传动电力机车系统集成国家重点实验室开放课题

Life prediction of lithium battery based on EEMD-GSGRU

Lingzhi YI1,3(), Zongguang ZHANG1,2(), Chaodong FAN1,2, Xianguang LUO4, Wang LI4, Wenhan LIU1,2   

  1. 1.Key Laboratory of Intelligent Computing & Information Processing (Xiangtan University), Xiangtan 411105, Hunan, China
    2.Hunan Province Engineering Research Center for Multi-Energy Collaborative Control Technology, Xiangtan 411105, Hunan, China
    3.Hunan Province Cooperative Innovation Center for Wind Power Equipment and Energy Conversion, Xiangtan 411105, Hunan, China
    4.The State Key Laboratory of Heavy Duty AC Drive Electric Locomotive Systems Integration, Zhuzhou 412001, Hunan, China
  • Received:2020-01-08 Revised:2020-05-25 Online:2020-09-05 Published:2020-09-08
  • Contact: Lingzhi YI E-mail:ylzwyh@xtu.edu.cn;1574511421@qq.com

摘要:

针对构建锂电池寿命预测数学模型复杂且易出现过拟合及泛化能力差的问题,提出集合经验模态分解(EEMD)和门控循环单元(GRU)并采用网格搜索(GS)的时序分解-集成模型(EEMD-GSGRU)。该模型首先将锂电池剩余容量数据进行信号分解,分解成总占比大的趋势因子和总占比小的误差因子,将分解时间序列分别预测GRU再合并进行实时滚动预测,最后使用GS搜索网络参数,使用Adam优化策略更新GRU网络权重。最后将NASA提供的锂电池数据集用于该模型,并与其他算法进行对比证明本EEMD-GSGRU模型优越性,本模型10次实验的平均均方根误差分别为0.0169、0.0309、0.0111,平均绝对百分误差分别为1.1921、2.2706、0.7279,证明本模型提高了锂电池寿命预测精度。

关键词: 锂电池, 寿命预测, 信号分解, 门控循环网络, 预测精度

Abstract:

Mathematical models for predicting lithium battery lives are complex, prone to over-fitting, and have poor generalization; accordingly, a time series decomposition-integration model (EEMD-GSGRU) based on Ensemble Empirical Mode Decomposition (EEMD) and Gated Recurrent Unit (GRU) with Grid Search (GS) is proposed. In this model, the lithium battery capacity data are first decomposed into the trend factor with a large total proportion and the error factor with a small total proportion. Then, the decomposed time series are predicted for GRU and combined for a real-time rolling prediction. Finally, GS is used to search the network parameters and Adam optimization is used to update the network weight of the GRU. Using a lithium battery dataset provided by NASA for the model, the superiority of the EEMD-GSGRU model is proved in comparison to other algorithms. It is shown that the EEMD-GSGRU model improves the accuracy of the lithium battery life prediction.

Key words: lithium battery, life prediction, signal decomposition, gated recurrent unit, prediction accuracy

中图分类号: