摘要
准确估计锂离子电池的健康状态(SOH)对于推动电动汽车的发展至关重要。然而,复杂的电池老化机制仍然是主要挑战。为此,提出了一种基于混合特征提取和机器学习的SOH估计模型。首先,基于增量容量分析(ICA)和Spearman相关性分析,从经过滑动平均滤波的增量容量(IC)曲线提取特征。其次,从电化学阻抗谱的2种可视化表示(Nyquist图和Bode图)中提取特征。然后,利用k均值聚类方法精简上述特征,并建立基于贝叶斯优化的CNN-BiGRU-Attention网络的SOH估计模型。进一步,通过网络中的1维卷积挖掘原始特征中的深层次信息,并添加了多头注意力机制的BiGRU网络,能够更有效地捕获输入序列中包含电池老化的关键信息,从而估计电池SOH。最后,利用3个不同电极材料的电池数据集进行实验,验证了所提方法的有效性。
Accurately estimating the state of health(SOH)of lithium-ion batteries is crucial for promoting the development of electric vehicles.However,the complex battery aging mechanism remains a major challenge.Therefore,an SOH estimation model based on hybrid feature extraction and machine learning is proposed.Firstly,based on incremental capacity analysis(ICA)and Spearman correlation analysis,features are extracted from the incremental capacity(IC)curve after sliding average filtering.Secondly,features are extracted from two visual representations of electrochemical impedance spectroscopy,namely Nyquist plot and Bode plot.Then,the k-means clustering method is used to streamline the above features,and an SOH estimation model is established for the CNN-BiGRU-Attention network based on Bayesian optimization.Furthermore,one-dimensional convolution in the network is used to excavate the deep-level information in the original features.And the bi-directional gated recurrent unit(BiGRU)network with multi-head attention mechanism can capture the key information containing battery aging in the input sequence more effectively to estimate SOH of batteries.Finally,experiments are conducted using battery datasets with three different electrode materials to validate the effectiveness of the proposed method.
作者
谢国民
刘澳
XIE Guomin;LIU Ao(Faculty of Electrical and Control Engineering,Liaoning Technical University,Huludao 125105,China)
出处
《电力系统自动化》
北大核心
2025年第21期120-130,共11页
Automation of Electric Power Systems
基金
国家自然科学基金资助项目(51974151)
辽宁省教育厅基础项目(LJKMZ20220683)。
关键词
锂离子电池
健康状态
特征提取
机器学习
增量容量分析
卷积神经网络(CNN)
双向门控循环单元(BiGRU)
多头注意力机制
lithium-ion battery
state of health(SOH)
feature extraction
machine learning
incremental capacity analysis
convolutional neural network(CNN)
bi-directional gated recurrent unit(BiGRU)
multi-head attention mechanism
