储能科学与技术 ›› 2020, Vol. 9 ›› Issue (1): 138-144.doi: 10.19799/j.cnki.2095-4239.2019.0190

• 储能系统与工程 • 上一篇    下一篇

基于ACO-BP神经网络的锂离子电池容量衰退预测

张新锋1,2(), 姚蒙蒙1,2, 王钟毅1,2, 饶勇翔1,2   

  1. 1. 长安大学汽车运输安全保障技术交通行业重点实验室
    2. 长安大学汽车学院，陕西 西安 710064
  • 收稿日期:2019-08-27 修回日期:2019-09-16 出版日期:2020-01-05 发布日期:2019-09-27
  • 作者简介:联系人：张新锋(1976—),男，博士，副教授，硕士生导师，主要研究领域为智能网联汽车与交通，车辆可靠性理论与技术,E-mail:zhxf@chd.edu.cn。
  • 基金资助:
    中央高校基本科研业务费专项资金(CHD2012JC048);长安大学基础研究支持计划专向基金和汽车运输安全保障技术交通行业重点实验室开放基金资助

Lithium-ion battery capacity decline prediction based on ant colony optimization BP neural network algorithm

ZHANG Xinfeng1,2(), YAO Mengmeng1,2, WANG Zhongyi1,2, RAO Yongxiang1,2   

  1. 1. Key Laboratory of Automotive Transportation Safety and Security Technology Transportation Industry, Chang'an University
    2. College of Automobile, Chang'an University, Xi'an 710064, Shaanxi, China
  • Received:2019-08-27 Revised:2019-09-16 Online:2020-01-05 Published:2019-09-27

RichHTML

14

PDF

320

摘要：

准确预测电池的容量衰退趋势对加强电池系统的管理和维护具有重要意义。本工作选择以锂离子电池为研究对象，根据NASA实验室公开的源数据集分析、预测锂离子电池的容量衰退趋势。在室温、恒流工况下对锂离子电池进行满充满放的循环充放电试验，得到各循环周期下电池的实际额定容量值，采用紧支集正交小波分析对获得的电池监测数据进行去噪优化处理，得到更加平稳规律的电池容量衰退过程，然后利用蚁群算法（Ant colony optimization，ACO）优化BP神经网络的初始权值和阈值，基于ACO-BP神经网络模型完成对锂离子电池容量衰退的预测，并与单独使用BP神经网络进行对比。结果表明，采用ACO-BP神经网络比单独使用BP神经网络具有更好的预测效果，且随着训练样本的增加，包含更多的电池容量退化信息，预测精度明显提高，当以前80个循环充放电周期作为训练样本时，预测的平均误差为1.46%，若继续扩大训练样本，预测效果将会更好。本研究有助于加强电池系统的健康管理，为高效预测锂离子电池的劣化轨迹提供技术参考。

关键词: 锂离子电池, 容量衰退, 小波分析, 蚁群算法, BP神经网络

Abstract:

Accurately predicting the declining trend with respect to the capacity of a battery is important for strengthening the management and maintenance of the battery system. Lithium-ion batteries are the research object; the battery capacity decline trend is predicted based on a source data set analysis published by NASA Laboratories. The data for a full-cycle charge-discharge test of a battery, obtained at room temperature and constant current, are denoised and optimized by a compact set orthogonal wavelet analysis to obtain a more stable and regular battery capacity decay process. The ant colony optimization (ACO) algorithm is subsequently used to optimize the initial weight of the BP neural network. And threshold, based on the ACO-BP neural network model to predict the capacity decline of lithium-ion batteries, and compared with BP neural network alone. The results denote that the ACO-BP neural network generates better prediction results when compared with that generated by the BP neural network alone; with more training samples, it contains more information on battery capacity degradation, and the prediction accuracy is significantly improved. The predicted average error is 1.46% when 80 charge and discharge cycles are used as training samples. If the training samples are further expanded, the prediction effect will improve. This study helps to strengthen the management of the battery systems and provides a technical reference for efficiently predicting the degradation trajectory of the lithium-ion batteries.

Key words: lithium-ion battery, capacity decline, wavelet analysis, ant colony algorithm, BP neural network

中图分类号: