电子学报 ›› 2022, Vol. 50 ›› Issue (3): 619-624.DOI: 10.12263/DZXB.20210207

• 学术论文 • 上一篇    下一篇

多状态影响下基于Bi‑LSTM网络的锂电池剩余寿命预测方法

张浩, 胡昌华, 杜党波, 裴洪, 张建勋   

  1. 火箭军工程大学导弹工程学院, 陕西 西安 710025
  • 收稿日期:2021-02-02 修回日期:2021-06-01 出版日期:2022-03-25 发布日期:2022-03-25
  • 作者简介:张 浩 男,1996年生于陕西安康.现为火箭军工程大学导弹工程学院硕士研究生.主要研究方向为基于深度学习的设备剩余寿命估计与故障预测.E‑mail:zhzshjj@163.com
    胡昌华 男,1966年生于湖北罗田.现为火箭军工程大学教授、博士生导师.主要研究方向为复杂系统故障诊断,寿命预测与容错控制.E‑mail:hch66603@163.com
    杜党波 男,1989年生于陕西西安.现为火箭军工程大学讲师.主要研究方向为预测与健康管理、 剩余寿命估计.E‑mail:luyq@mail.tsinghua.edu.cn
    裴 洪 男,1992年生于安徽六安.现为火箭军工程大学讲师.主要研究方向为装备的剩余寿命预测与维修决策.E‑mail:ddb_effort@126.com
    张建勋 男,1988年生于四川南充.现为火箭军工程大学讲师.主要研究方向为预测与健康管理、退化过程建模、剩余寿命估计.E‑mail:zhang200735@163.com
  • 基金资助:
    国家自然科学基金(61833016);国家重点研发项目(2018YFB1306100);陕西省自然科学基金(2020JQ-489)

Remaining Useful Life Prediction Method of Lithium‑Ion Battery Based on Bi‑LSTM Network Under Multi‑State Influence

ZHANG Hao, HU Chang-hua, DU Dang-bo, PEI Hong, ZHANG Jian-xun   

  1. College of Missile Engineering,Rocket Force University of Engineering,Xi’an,Shaanxi 710025,China
  • Received:2021-02-02 Revised:2021-06-01 Online:2022-03-25 Published:2022-03-25

摘要:

现有基于深度学习的锂电池剩余寿命(Remaining Useful Life, RUL)预测方法中,锂电池多个内部状态所蕴含的寿命信息未得到充分考虑.鉴于此,提出了一种融合电池容量、阻抗与温度三个内部状态的RUL预测模型.首先,引入双向长短时记忆(Bi?directional Long Short?Term Memory, Bi?LSTM)网络学习三种状态数据的时间相关性.其次,利用dropout技术与Bayesian变分推断技术间的等价性实现了RUL预测结果的不确定性量化,得到了预测结果的95%置信区间与概率密度分布(Probability Density Function, PDF),并分析了不同dropout率对预测不确定性的影响.最后,通过四种不同的深度学习模型框架与两种内部状态输入方案的对比实验,验证了本文方法的有效性.

关键词: 深度学习, 剩余寿命预测, Bi?LSTM网络, Bayesian变分推断技术, dropout技术, 不确定性量化

Abstract:

The life information contained in multiple internal states of lithium-ion battery is not fully considered in the existing RUL(Remaining Useful Life) prediction methods of lithium-ion battery based on deep learning. In view of this, a RUL prediction model that integrates the three internal states include battery capacity, impedance and temperature is proposed. The Bi-LSTM(Bi-directional Long Short-Term Memory) network is introduced to learn the correlation about time of the data of the three states firstly. Secondly, the equivalence between dropout technology and Bayesian variational inference technology is used to quantify the uncertainty of the RUL prediction results. The 95% confidence interval and PDF(Probability Distribution Function) of the RUL prediction results are obtained, and the effect on the prediction uncertainty of different dropout rates is analyzed. Finally, the effectiveness of this method is verified through the comparative experiments of four different deep learning model and two input schemes of internal state.

Key words: deep learning, remaining useful life prediction, bi-directional long short-term memory network, Bayesian variational inference technology, dropout technology, quantification of uncertainty

中图分类号: