Domain generalization-based state-of-health estimation of lithium-ion batteries

• Published in: Journal of power sources 2024-08, Vol.610, p.234696, Article 234696
• Main Authors: Chen, Liping, Bao, Xinyuan, Lopes, António M., Li, Xin, Kong, Huifang, Chai, Yi, Li, Penghua
• Format: Article
• Language: English
• Subjects: Convolutional neural network; Gated recurrent neural network; Lithium-ion batteries; State-of-health estimation
• ISSN: 0378-7753
• DOI: 10.1016/j.jpowsour.2024.234696

This paper proposes a domain generalization-based method for state-of-health (SOH) estimation of lithium-ion batteries. First, a combination of convolutional neural network and gated recurrent neural network is used to obtain features from input data in different domains. Second, a SOH estimator is used to estimate the SOH and to obtain the estimated loss. Then, a domain discriminator is used to classify the pairwise domains, while the domain classification loss is calculated. After that, to increase the diversity, all input data from different domains are taken as a whole, to obtain the estimated value and its corresponding loss, through the feature extractor and SOH estimator. Finally, all calculated losses are added, the result is treated as the loss function of the network, and each parameter of the model is optimized through back-propagation. Experimental results show that the proposed method has better performance than various state-of-the-art baselines approaches. •A lithium-ion battery SOH estimation method based on DG is proposed.•A model is trained on a known domain to achieve SOH prediction on an unknown domain.•Mix multiple source domains together to form new domains to increase data diversity.•The proposed method has better performance compared with many advanced methods.