文献类型：期刊文献

英文题名：Bidirectional-Mamba-based iterative prediction for long-cycle battery remaining useful life with physical constraints

作者：Gao, Zibo[1];Zhuang, Zixian[1];Luan, Weiling[1];Chen, Ying[1]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Key Lab Adv Battery Syst & Safety CPCIF, Shanghai 200237, Peoples R China

年份：2026

卷号：141

外文期刊名：JOURNAL OF ENERGY STORAGE

收录：;EI(收录号：20254419436336);WOS:【SCI-EXPANDED(收录号:WOS:001612012200001)】；

基金：The authors gratefully acknowledge the support from the National Natural Science Foundation of China (52375144 and 52205153) , Shanghai Pujiang Programme (23PJD019) and the Fundamental Research Funds for the Central Universities (JKG01251734) during the course of this work.

语种：英文

外文关键词：Lithium-ion battery; Remaining useful life; Bi-Mamba; Physical constraints; Whale optimization algorithm

摘要：Accurate long-cycle remaining useful life (RUL) prediction is critical for improving battery safety and reducing maintenance costs in electric vehicles and energy storage systems. A new iterative deep learning model based on bidirectional Mamba (Bi-Mamba) and physical constraints is proposed, which is optimized by the whale optimization algorithm (WOA), and only the early capacity data is used to predict the RUL of long-cycle battery. By integrating a denoising autoencoder (DAE) and positional encoding, the proposed model achieves enhanced robustness against noise and improved learning of temporal dependencies in battery capacity degradation trajectories. To effectively mitigate cumulative errors inherent in long-sequence forecasts, the framework applies a ten steps iterative prediction strategy. Validation on two public battery datasets (CALCE and MIT-Stanford) demonstrates that the proposed method attains significantly lower prediction errors (average relative errors of 1.60 % and 0.67 %, respectively) compared to advanced benchmark models including bidirectional long short-term memory (Bi-LSTM), bidirectional gated recurrent unit (Bi-GRU), and DeTransformer, especially in scenarios exhibiting capacity fluctuations phenomena. Experiments on different prediction starting points show that the model has good robustness, and the prediction accuracy gradually improves with the backward shift of the prediction starting point. The proposed approach highlights a practical and effective solution suitable for real-time deployment in battery management systems (BMS), laying foundations for further advancements in multi-sensor data fusion and embedded predictive maintenance implementations.