| 摘要: |
| 目的 健康状态是评估锂离子电池状态的关键参数,对锂离子电池的安全使用有着十分重要的意义,为了获
得准确可靠的健康状态估计结果,建立基于卷积神经网络和 Transformer 的锂离子电池健康状态估计方法,利用不
同模型的数据挖掘特性,将健康指标的深层信息和随循环周期增加的时序信息并行提取。 方法 从锂离子电池放电
过程中的部分电压和温度曲线中提取 3 个与健康状态相关性较强的健康指标作为模型输入,利用卷积神经网络强
大的特征提取能力挖掘健康指标的局部特征,利用 Transformer 的顺序处理能力挖掘健康指标的时序特征,将健康
指标的局部特征和时序特征进行特征融合,通过卷积和全局平均池化层输出健康状态估计值。 结果 本研究使用
MIT 数据集进行实验验证,并与卷积神经网络和长短时记忆神经网络进行对比分析,所提出的方法的均方根误差
和平均绝对误差是最低的,为 0. 11 和 0. 08,最小相对误差为 0. 61%。 结论 所提出的 CNN-Transformer 健康状态估
计采用不同模型挖掘健康指标不同的特征信息,能够充分利用锂离子电池放电数据,且具有良好的估计效果。 |
| 关键词: 锂离子电池 健康状态 卷积神经网络 Transformer |
| DOI： |
| 分类号: |
| 基金项目: |
|
| State of Health Estimation of Lithium-ion Battery Based on CNN-Transformer |
|
CHEN Guolin, YAO Xingyan
|
|
School of Artificial Intelligence, Chongqing Technology and Business University, Chongqing 400067, China
|
| Abstract: |
| Objective The state of health is a key parameter to evaluate the state of lithium-ion batteries and it is of great
significance to the safe use of lithium-ion batteries. To obtain accurate and reliable health state estimation results a health
state estimation method for lithium-ion batteries based on convolutional neural networks and Transformer was developed.
Using the data mining characteristics of different models the deep information of health indicators and the time series
information that increases with the cycle period were extracted in parallel. Methods Three health indicators with a strong
correlation with health status were extracted from the partial voltage and temperature curves of the lithium-ion battery
during the discharge process as model inputs. The powerful feature extraction capability of the convolutional neural
network was used to mine the local features of health indicators and the sequential processing capability of the
Transformer was used to mine the time series features of health indicators. The local and time-series features of health
indicators were fused with features and the health status estimates were output by convolution and global average pooling layers. Results In this study experimental validation was performed using the MIT dataset a comparative analysis was
performed with convolutional neural network and long and short-term memory neural network and the root mean square
error and mean absolute error of the proposed method were the lowest which were 0. 11 and 0. 08 and the minimum
relative error was 0. 61%. Conclusion The experimental results show that the proposed CNN-Transformer health state
estimation uses different models to mine different feature information of health indicators which can make full use of
lithium-ion battery discharge data and has a good estimation effect. |
| Key words: lithium-ion battery state of health convolutional neural network Transformer |
