| 引用本文: | 王一航,冯良骏,赵春晖.面向锂电池少量循环的二维支持域直推式健康状态预测[J].控制理论与应用,2024,41(3):474~483.[点击复制] |
| WANG Yi-hang,FENG Liang-jun,ZHAO Chun-hui.Lithium battery two-dimensional region of support transductive learning and state of health prediction oriented to few charge-discharge cycles[J].Control Theory and Technology,2024,41(3):474~483.[点击复制] |
|
| 面向锂电池少量循环的二维支持域直推式健康状态预测 |
| Lithium battery two-dimensional region of support transductive learning and state of health prediction oriented to few charge-discharge cycles |
| 摘要点击 2479 全文点击 72 投稿时间：2022-05-18 修订日期：2024-02-26 |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| DOI编号 10.7641/CTA.2023.20408 |
| 2024,41(3):474-483 |
| 中文关键词 锂电池健康状态 少量充放电循环 二维支持域 直推式学习 |
| 英文关键词 lithium battery SOH few charge-discharge cycles 2-D region of support transductive learning |
| 基金项目 国家杰出青年科学基金项目(62125306), 国家自然科学基金重点项目(62133003), 工业控制技术国家重点实验室自主课题项目(ICT2021A15)资助. |
|
| 中文摘要 |
| 锂离子电池的健康状态(SOH)是反映电池老化程度的关键指标, 但由于电池老化的非线性和不确定性使得
SOH难以精确估计, 并且受到电池数据收集的高时间成本和容量再生现象的影响, 传统的数据驱动方法在历史充放
电循环数较少时效果较差. 针对上述问题, 本文创新性地提出了一种二维支持域直推式学习(2D-RoSTL)建模思路,
建立了数据由粗到细的精准划分方法, 用于少量充放电循环下的SOH预测. 一方面, 考虑同型号多块电池的批次特
性, 利用历史数据和批次数据构造二维支持域扩充模型信息来源, 提供了粗范围的大量可供选择的样本; 另一方面,
首次尝试以直推式学习的方式解决SOH预测的任务, 利用离线和在线样本特征空间的信息, 对每个样本进行细致划
分, 提升少量充放电循环情况下模型的预测可靠性. 基于NASA的公开数据集, 所提出的二维支持域直推式建模方
法在4个电池上的预测误差均小于1.56%, 实现了对锂电池充放电历史初期及再生点的精确预测. |
| 英文摘要 |
| The state of health (SOH) of lithium-ion batteries is a key indicator reflecting the degree of battery aging.
However, due to the nonlinearity and uncertainty of battery aging, it is challenging to estimate SOH accurately. Besides,
affected by the high time cost of battery data collection and the phenomenon of capacity regeneration, the traditional
data driving method is less effective when the number of historical charge-discharge cycles is small. To solve the above
problems, this paper innovatively proposes a two-dimensional region of support transductive learning (2D-RoSTL) method,
and establishes a precise data division method from coarse to fine, which is used for SOH prediction oriented to few chargedischarge
cycles. On the one hand, considering the batch characteristics of the same model block battery, using historical
data and batch data to construct a two-dimensional region of support to expand the information source of the model,
providing an extensive range of samples for selection. On the other hand, we first attempt to solve the SOH prediction task
by transductive learning method. Using offline and online information in the sample feature space, our model finely divides
each sample to improve the online prediction reliability in the case of a few charge-discharge cycles. Based on the NASA’s
public data set, the proposed two-dimensional region of support transductive modeling method has a prediction error of less
than 1.56% on the four batteries, and has realized the accurate prediction of the early history of lithium batteries and the
point of regeneration. |
|
|
|
|
|