State-of-health estimation for lithium-ion batteries using unsupervised deep subdomain adaptation

Feifan Li, Yongguang Yu, Xiaolin Yuan and Guojian Ren

Energy, 2025, vol. 324, issue C

Abstract: In battery management systems (BMS), it is always a challenge to perform cross-domain state of health (SOH) estimation among different lithium-ion batteries (LIBs). To address this problem, this paper proposes an unsupervised transfer learning framework based on subdomain adaptation aiming at accurate SOH prediction. Specifically, a deep subdomain transfer network (DSTN) is constructed by integrating a convolutional neural network (CNN) with a long-short-term memory (LSTM) network and incorporating a multi-head attention mechanism (MHA) in order to deeply mine and merge multi-source feature information. This mechanism effectively enhances the ability of the model to capture key information from the complex feature space, and relies on the unsupervised learning to extract effective features for SOH prediction. For the case that the labeled data belongs to the continuous domain, the local maximum mean difference (LMMD) is employed to reduce the distributional difference between the source and target domain battery data by using the subdomain adaptive theory to achieve the leap from the classification task to the regression task. In the experimental part, the significant advantages of LMMD in measuring inter-domain differences are highlighted by systematically comparing LMMD with the traditional maximum mean difference (MMD) method. In addition, a comparative experiment is carried out between adding and not adding a multi-head attention layer in the network. It not only verifies the accuracy and efficiency of DSTN in transfer learning tasks, but also highlights the key role of multiple attention mechanism in enhancing transfer learning tasks. Furthermore, through comparison with supervised learning scenarios, it has been confirmed that the unsupervised LMMD method can demonstrate excellent performance even in practical applications where sufficient labeled data are lacking. This highlights its broad applicability and practicality. The experimental results show that the unsupervised transfer learning method based on lmmd in this paper can improve the accuracy of transfer learning and has wide popularization significance.

Keywords: Lithium-ion batteries (LIBs); Unsupervised transfer learning; Local maximum mean difference (LMMD); Multi-head attention (MHA); State of health (SOH) estimation (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S036054422501504X
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:energy:v:324:y:2025:i:c:s036054422501504x

DOI: 10.1016/j.energy.2025.135862

Access Statistics for this article

Energy is currently edited by Henrik Lund and Mark J. Kaiser

More articles in Energy from Elsevier
Bibliographic data for series maintained by Catherine Liu ().