Scholarlink

<p>The state-of-health (SOH) estimation of lithium-ion batteries is significantly constrained by the heterogeneity of temporal evolution and the collinear redundancy inherent in multivariate data. To address these challenges, this study proposes a Bi-MF (Bi-Mamba + &amp; Transformer) model founded on a spatiotemporal multi-dimensional decoupling strategy. In the temporal dimension, the model integrates Bi-Mamba + with multi-scale dilated attention (MSDA) to preserve long-term degradation memory throughout the full life cycle while capturing local nonlinear fluctuations, such as capacity regeneration. In the sequence dimension, the collaboration between a series-relation-aware (SRA) decision-maker and a channel aggregation (CA) mechanism adaptively mitigates noise interference within highly coupled features. Experimental results on the NASA and MIT datasets demonstrate that the Bi-MF model effectively overcomes the issues of long-sequence prediction divergence and local lag, exhibiting superior robustness and accuracy under high-rate and complex aging conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

State of health estimation for lithium-ion batteries based on a parallel architecture fusing Bi-Mamba + and MSDA

  • 期刊论文

  • 发表日期：

    2026年5月4日

查看全文
  • Jinhua Wang,
  • Yingwen Ma,
  • Jie Cao

摘要

The state-of-health (SOH) estimation of lithium-ion batteries is significantly constrained by the heterogeneity of temporal evolution and the collinear redundancy inherent in multivariate data. To address these challenges, this study proposes a Bi-MF (Bi-Mamba + & Transformer) model founded on a spatiotemporal multi-dimensional decoupling strategy. In the temporal dimension, the model integrates Bi-Mamba + with multi-scale dilated attention (MSDA) to preserve long-term degradation memory throughout the full life cycle while capturing local nonlinear fluctuations, such as capacity regeneration. In the sequence dimension, the collaboration between a series-relation-aware (SRA) decision-maker and a channel aggregation (CA) mechanism adaptively mitigates noise interference within highly coupled features. Experimental results on the NASA and MIT datasets demonstrate that the Bi-MF model effectively overcomes the issues of long-sequence prediction divergence and local lag, exhibiting superior robustness and accuracy under high-rate and complex aging conditions.