欢迎访问《汽车安全与节能学报》，
汽车安全与节能学报

汽车安全与节能学报 ›› 2022, Vol. 13 ›› Issue (3): 541-549.DOI: 10.3969/j.issn.1674-8484.2022.03.016

• 汽车节能与环保 • 上一篇    下一篇

基于dropout-MC递归神经网络的锂电池剩余寿命预测

魏孟1,2(), 王桥1, 叶敏1,*(), 廉高棨1, 徐信芯1,3   

  1. 1.长安大学 公路养护装备国家工程实验室,西安 710064, 中国
    2.新加坡国立大学 机械系,新加坡 117576, 新加坡
    3.河南省高等级公路检测与养护技术重点实验室, 新乡 453003, 中国
  • 收稿日期:2022-02-09 修回日期:2022-04-28 出版日期:2022-09-30 发布日期:2022-10-04
  • 通讯作者: 叶敏
  • 作者简介:* 叶敏,教授。E-mail: mingye@chd.edu.cn。
    魏孟（1997—）,男（汉）,陕西,博士研究生。E-mail: weimeng@chd.edu.cn。
  • 基金资助:
    陕西省科技创新团队(2020TD0012);陕西省青年科技新星项目(2020KJXX-044);中央高校基金优秀博士论文资助项目CHD(300203211251);河南省杰出外籍科学家工作室(GZS2022004)

Remaining useful life prediction of lithium-ion batteries based on dropout Monte Carlo recurrent neural network

WEI Meng1,2(), WANG Qiao1, YE Min1,*(), LIAN Gaoqi1, XU Xinxin1,3   

  1. 1. National Engineering Laboratory for Highway Maintenance Equipment, Chang ‘an University, Xi ‘an, 710064
    2. Department of Mechanical Engineering, National University of Singapore, Singapore 117576, Singapore
    3. Henan Key Laboratory of High-Grade Highway Detection and Maintenance Technology, Xinxiang 453003, China
  • Received:2022-02-09 Revised:2022-04-28 Online:2022-09-30 Published:2022-10-04
  • Contact: YE Min

RichHTML

2

PDF

76

摘要：

为保障电动车辆的可靠性和安全性,提出了一种dropout Monte Carlo（dropout-MC）递归神经网络的锂离子动力电池系统的剩余寿命（RUL）预测方法。以等电压充电时间间隔作为间接健康因子,考虑外部干扰和容量再生现象的影响,以变分模态分解（VMD）来获得电池退化趋势。以改进的递归神经网络模型——长短时间序列（LSTM）来获得剩余寿命预测。以dropout-MC采样方法来表征锂离子电池剩余寿命的不确定性,并获得锂离子电池RUL的95%置信区间。结果表明：相较于传统的极限学习机（ELM）方法和非线性自回归神经网络（NARX）方法,该文方法的剩余寿命预测性能指标均低于2.4%。因而,该方法具有优越的预测性能,且获得预测的置信区间。

关键词: 车辆安全, 锂离子电池, 剩余寿命（RUL）, 变分模态分解（VMD）, dropout Monte Carlo（dropout-MC）方法, 递归神经网络

Abstract:

A dropout Monte Carlo (dropout-MC) recurrent neural network method was proposed for remaining useful life (RUL) prediction for lithium-ion batteries to guarantee the safety and reliability of electric vehicles. An equal charging voltage time was introduced as an indirect health indicator, and the variational mode decomposition (VMD) was adopted to reduce the influence of external interference and capacity regeneration. The long and short time series (LSTM) as the improved recurrent neural network was established for accurate RUL prediction. The dropout-MC method was proposed to obtain the 95% confidence interval for quantifying the uncertainty of the RUL prediction. Compared with traditional extreme learning machine (ELM) and nonlinear autoregressive neural network (NARX) methods, the proposed method not only can achieve a higher accuracy in RUL prediction with prediction performance below 2.4%, but also obtain reliability of RUL prediction.

Key words: vehicle safety, lithium-ion batteries, remaining useful life (RUL), variational mode decomposition (VMD), dropout Monte Carlo (dropout-MC) method, recurrent neural network

中图分类号: