Lithium-ion batteries are an important energy storage sources, and it is of great practical importance to predict their remaining useful life (RUL). First, the battery data are treated as matrices, and singular value decomposition (SVD) is introduced to extract the potential health indicator (HI) from the measured data and the feature extraction objects containing more degradation information. This will address the drawbacks of traditional feature extraction methods that rely on parameter settings and poor adaptability to different lithium-ion battery datasets. Second, the redundancy and deficiency of potential HIs affect the prediction of RUL, and thus, a fused HI is obtained by processing HIs using Spearman correlation analysis and stacked autoencoder, considering the shortcomings of principal component analysis (PCA). Accordingly, a model between fused HI and capacity is constructed using the Gaussian process regression algorithm, and the final prediction results with uncertainty expression are obtained. Finally, the feasibility and validity of the proposed prediction model are verified by four aging batteries provided by NASA. The MIT battery dataset is used to verify the adaptability of the feature extraction method. The experimental results show that the proposed RUL prediction framework has good prediction performance and that the SVD feature extraction method has good adaptability while avoiding parameter settings. The HI extracted in this paper has significantly improved the prediction accuracy compared with the HI after PCA fusion and other HIs.
Keywords:lithium-ion batteries
;
remaining useful life
;
singular value decomposition
;
stacked autoencoder
;
gaussian process regression
DONG Yuanchang. Remaining useful life prediction of lithium-ion batteries based on SVD-SAE-GPR[J]. Energy Storage Science and Technology, 2023, 12(4): 1257-1267
基于以上的分析,本工作对电池的原始测量数据重新进行分析,尝试从中获取隐藏的特征信息。根据电池数据的特点,同时考虑到特征值是矩阵的重要信息,本工作将电池数据看作矩阵,从矩阵中计算特征值,然后判断该特征值是否可以作为重要信息来表征电池的退化状态。但电池数据不是方阵,无法获得其特征值,而奇异值分解[14](singular value decomposition,SVD)可以从非方阵中提取奇异值来表示矩阵的重要信息。因此,本文选择SVD从电池数据中计算奇异值作为HIs。同时,考虑到提取的HIs会有一定的冗余性,当作为训练数据时,其维度过高和不足会使模型训练变得复杂且不利于RUL预测的准确性。因此,本工作的第二个动机是消除HIs不相关信息的负面影响和降低数据维数。在数据去噪和降维方面,主成分分析[15](principal component analysis,PCA)表现出了较好的性能,但它适合于服从高斯分布的数据且只能执行线性变换,缺乏一定的灵活性。堆叠自编码器[16](stacked autoencoder,SAE)可以自动从无标注数据中提取学习特征,给出比原始数据更好的特征描述。因此,选择SAE用于HIs的在线融合。
此外,模型和数据的不确定性可能导致预测可靠性差,因此,在研究锂离子电池的RUL预测时考虑预测结果的不确定性是非常有必要的。高斯过程回归[17](gaussian process regression,GPR)能够实现任意线性或者非线性系统动态行为特征的建模预测,并能以概率的形式解释预测结果的不确定性。考虑到锂离子电池的退化过程是一个复杂的、动态的、非线性的电化学过程,GPR方法适合于建立锂离子电池RUL预测模型。
为了验证SVD特征提取方法对不同锂离子电池数据集的适应性,本小节的实验额外选择了麻省理工学院[23](Massachusetts Institute of Technology,MIT)提供的电池数据集,其中电池参数和实验环境与NASA提供的电池数据集不同。首先详细描述了用于测试的电池数据集。然后,仍然使用SVD提取HIs。最后,利用GPR进行容量预测。
LIANG X C, ZHANG M, HUANG G J. Review on lithium-ion battery modeling methods based on BMS[J]. Energy Storage Science and Technology, 2020, 9(6): 1933-1939.
HU X S, ZHANG K, LIU K L, et al. Advanced fault diagnosis for lithium-ion battery systems: A review of fault mechanisms, fault features, and diagnosis procedures[J]. IEEE Industrial Electronics Magazine, 2020, 14(3): 65-91.
CHEN Y, BAI Y F, HE Y. Comparison of data-driven lithium battery state of health estimation methods[J]. Energy Storage Science and Technology, 2019, 8(6): 1204-1210.
PANG X Q, ZHAO Z, WEN J, et al. An interval prediction approach based on fuzzy information granulation and linguistic description for remaining useful life of lithium-ion batteries[J]. Journal of Power Sources, 2022, 542: doi: 10.1016/j.jpowsour.2022.231750.
CHU A, ALLAM A, CORDOBA ARENAS A, et al. Stochastic capacity loss and remaining useful life models for lithium-ion batteries in plug-in hybrid electric vehicles[J]. Journal of Power Sources, 2020, 478: doi: 10.1016/i.jpowsour.2020.228991.
KHALEGHI S, HOSEN M S, KARIMI D, et al. Developing an online data-driven approach for prognostics and health management of lithium-ion batteries[J]. Applied Energy, 2022, 308: doi: 10.1016/j.apenergy.2021.118348.
JIA J F, YUAN S F, SHI Y H, et al. Improved sparrow search algorithm optimization deep extreme learning machine for lithium-ion battery state-of-health prediction[J]. iScience, 2022, 25(4): doi:10.1016/j.isci.2022.103988.
LI X D, ZHANG X W. State of health estimation method for lithium-ion batteries based on principal component analysis and whale optimization algorithm-Elman model[J]. Energy Storage Science and Technology, 2022, 11(12): 4010-4021.
LIU J, CHEN Z Q. Remaining useful life prediction of lithium-ion batteries based on health indicator and Gaussian process regression model[J]. IEEE Access, 2019, 7: 39474-39484.
PANG X Q, LIU X Y, JIA J F, et al. A lithium-ion battery remaining useful life prediction method based on the incremental capacity analysis and Gaussian process regression[J]. Microelectronics Reliability, 2021, 127: doi: 10.1016/j.microrel.2021.114405.
PAN W J, LUO X S, ZHU M T, et al. A health indicator extraction and optimization for capacity estimation of Li-ion battery using incremental capacity curves[J]. Journal of Energy Storage, 2021, 42: doi: 10.1016/j.est.2021.103072.
WALL M E, RECHTSTEINER A, ROCHA L M. Singular value decomposition and principal component analysis[M]//A Practical Approach to Microarray Data Analysis. Boston: Kluwer Academic Publishers, 2005: 91-109.
GU X Y, SEE K W, LI P H, et al. A novel state-of-health estimation for the lithium-ion battery using a convolutional neural network and transformer model[J]. Energy, 2023, 262: doi:10.1016/j.energy.2022.125501.
WEI M, YE M, WANG Q, et al. Remaining useful life prediction of lithium-ion batteries based on stacked autoencoder and Gaussian mixture regression[J]. Journal of Energy Storage, 2022, 47: doi: 10.1016/j.est.2021.103558.
WANG R J, HUI Z L, YANG M. Gaussian process regression based on indirect health indicators for SOH estimation of lithium battery[J]. Energy Storage Science and Technology, 2023, 12(2): 560-569.
TANG T, YUAN H M. The capacity prediction of Li-ion batteries based on a new feature extraction technique and an improved extreme learning machine algorithm[J]. Journal of Power Sources, 2021, 514: doi: 10.1016/j.jpowsour.2021. 230572.
FLY A, CHEN R. Rate dependency of incremental capacity analysis (dQ/dV) as a diagnostic tool for lithium-ion batteries[J]. Journal of Energy Storage, 2020, 29:doi:10.1016/j.est.2020. 101329.
BLOOM I, CHRISTOPHERSEN J P, ABRAHAM D P, et al. Differential voltage analyses of high-power lithium-ion cells[J]. Journal of Power Sources, 2006, 157(1): 537-542.
SHIBAGAKI T, MERLA Y, OFFER G J. Tracking degradation in lithium iron phosphate batteries using differential thermal voltammetry[J]. Journal of Power Sources, 2018, 374: 188-195.
SEVERSON K A, ATTIA P M, JIN N, et al. Data-driven prediction of battery cycle life before capacity degradation[J]. Nature Energy, 2019, 4(5): 383-391.
CLERICI D, MOCERA F, SOMÀ A. Electrochemical-mechanical multi-scale model and validation with thickness change measurements in prismatic lithium-ion batteries[J]. Journal of Power Sources, 2022, 542: doi: 10.1016/j.jpowsour.2022. 231735.
ATTIA P M, GROVER A, JIN N, et al. Closed-loop optimization of fast-charging protocols for batteries with machine learning[J]. Nature, 2020, 578(7795): 397-402.
... 基于以上的分析,本工作对电池的原始测量数据重新进行分析,尝试从中获取隐藏的特征信息.根据电池数据的特点,同时考虑到特征值是矩阵的重要信息,本工作将电池数据看作矩阵,从矩阵中计算特征值,然后判断该特征值是否可以作为重要信息来表征电池的退化状态.但电池数据不是方阵,无法获得其特征值,而奇异值分解[14](singular value decomposition,SVD)可以从非方阵中提取奇异值来表示矩阵的重要信息.因此,本文选择SVD从电池数据中计算奇异值作为HIs.同时,考虑到提取的HIs会有一定的冗余性,当作为训练数据时,其维度过高和不足会使模型训练变得复杂且不利于RUL预测的准确性.因此,本工作的第二个动机是消除HIs不相关信息的负面影响和降低数据维数.在数据去噪和降维方面,主成分分析[15](principal component analysis,PCA)表现出了较好的性能,但它适合于服从高斯分布的数据且只能执行线性变换,缺乏一定的灵活性.堆叠自编码器[16](stacked autoencoder,SAE)可以自动从无标注数据中提取学习特征,给出比原始数据更好的特征描述.因此,选择SAE用于HIs的在线融合. ...
1
... 基于以上的分析,本工作对电池的原始测量数据重新进行分析,尝试从中获取隐藏的特征信息.根据电池数据的特点,同时考虑到特征值是矩阵的重要信息,本工作将电池数据看作矩阵,从矩阵中计算特征值,然后判断该特征值是否可以作为重要信息来表征电池的退化状态.但电池数据不是方阵,无法获得其特征值,而奇异值分解[14](singular value decomposition,SVD)可以从非方阵中提取奇异值来表示矩阵的重要信息.因此,本文选择SVD从电池数据中计算奇异值作为HIs.同时,考虑到提取的HIs会有一定的冗余性,当作为训练数据时,其维度过高和不足会使模型训练变得复杂且不利于RUL预测的准确性.因此,本工作的第二个动机是消除HIs不相关信息的负面影响和降低数据维数.在数据去噪和降维方面,主成分分析[15](principal component analysis,PCA)表现出了较好的性能,但它适合于服从高斯分布的数据且只能执行线性变换,缺乏一定的灵活性.堆叠自编码器[16](stacked autoencoder,SAE)可以自动从无标注数据中提取学习特征,给出比原始数据更好的特征描述.因此,选择SAE用于HIs的在线融合. ...
1
... 基于以上的分析,本工作对电池的原始测量数据重新进行分析,尝试从中获取隐藏的特征信息.根据电池数据的特点,同时考虑到特征值是矩阵的重要信息,本工作将电池数据看作矩阵,从矩阵中计算特征值,然后判断该特征值是否可以作为重要信息来表征电池的退化状态.但电池数据不是方阵,无法获得其特征值,而奇异值分解[14](singular value decomposition,SVD)可以从非方阵中提取奇异值来表示矩阵的重要信息.因此,本文选择SVD从电池数据中计算奇异值作为HIs.同时,考虑到提取的HIs会有一定的冗余性,当作为训练数据时,其维度过高和不足会使模型训练变得复杂且不利于RUL预测的准确性.因此,本工作的第二个动机是消除HIs不相关信息的负面影响和降低数据维数.在数据去噪和降维方面,主成分分析[15](principal component analysis,PCA)表现出了较好的性能,但它适合于服从高斯分布的数据且只能执行线性变换,缺乏一定的灵活性.堆叠自编码器[16](stacked autoencoder,SAE)可以自动从无标注数据中提取学习特征,给出比原始数据更好的特征描述.因此,选择SAE用于HIs的在线融合. ...
1
... 此外,模型和数据的不确定性可能导致预测可靠性差,因此,在研究锂离子电池的RUL预测时考虑预测结果的不确定性是非常有必要的.高斯过程回归[17](gaussian process regression,GPR)能够实现任意线性或者非线性系统动态行为特征的建模预测,并能以概率的形式解释预测结果的不确定性.考虑到锂离子电池的退化过程是一个复杂的、动态的、非线性的电化学过程,GPR方法适合于建立锂离子电池RUL预测模型. ...
1
... 此外,模型和数据的不确定性可能导致预测可靠性差,因此,在研究锂离子电池的RUL预测时考虑预测结果的不确定性是非常有必要的.高斯过程回归[17](gaussian process regression,GPR)能够实现任意线性或者非线性系统动态行为特征的建模预测,并能以概率的形式解释预测结果的不确定性.考虑到锂离子电池的退化过程是一个复杂的、动态的、非线性的电化学过程,GPR方法适合于建立锂离子电池RUL预测模型. ...
... 为了验证SVD特征提取方法对不同锂离子电池数据集的适应性,本小节的实验额外选择了麻省理工学院[23](Massachusetts Institute of Technology,MIT)提供的电池数据集,其中电池参数和实验环境与NASA提供的电池数据集不同.首先详细描述了用于测试的电池数据集.然后,仍然使用SVD提取HIs.最后,利用GPR进行容量预测. ...