Scholarlink

<p>Accurate remaining useful life (RUL) forecasting for lithium-ion batteries plays a crucial role in systems that rely on battery performance and is essential for ensuring operational safety and improving maintenance efficiency. However, due to the complex nonlinear degradation characteristics of batteries and the difficulty in modeling long-term dependencies, RUL prediction remains a challenging task. This study proposes a hybrid prediction framework based on Interactive Enhanced Mamba (IDMamba) and Efficient Compressed Sensing (ECS). ECS compresses the input sequence through a learnable Weibull measurement matrix, reducing redundancy while preserving key features. The dynamic FISTA algorithm with attention and normalization further enhances the reconstruction of effective information. IDMamba integrates cross-convolutional blocks and state-space modeling, enhanced by Dynamic Exponential Moving Average (DEMA). This adaptive smoothing module dynamically adjusts weights based on changes over time, achieving noise reduction while preserving local instantaneous dynamic changes. Extensive experiments on the CACLE and NASA datasets demonstrate the superiority of the proposed model in terms of RUL. Experimental results show that this method achieves higher prediction accuracy on multiple sets of battery data, while maintaining good robustness and generalization performance in cross-sample tests. Its overall performance is superior to existing comparative models.</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

Lithium-ion battery remaining useful life prediction based on efficient compressed sensing and interactive enhanced Mamba model

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Hai-Kun Wang,
  • Xin Liu,
  • Kaitao Chen,
  • Zhi-Chao Xu,
  • Yu-Kai Guo,
  • Qian Huang,
  • Huamei Cao

摘要

Accurate remaining useful life (RUL) forecasting for lithium-ion batteries plays a crucial role in systems that rely on battery performance and is essential for ensuring operational safety and improving maintenance efficiency. However, due to the complex nonlinear degradation characteristics of batteries and the difficulty in modeling long-term dependencies, RUL prediction remains a challenging task. This study proposes a hybrid prediction framework based on Interactive Enhanced Mamba (IDMamba) and Efficient Compressed Sensing (ECS). ECS compresses the input sequence through a learnable Weibull measurement matrix, reducing redundancy while preserving key features. The dynamic FISTA algorithm with attention and normalization further enhances the reconstruction of effective information. IDMamba integrates cross-convolutional blocks and state-space modeling, enhanced by Dynamic Exponential Moving Average (DEMA). This adaptive smoothing module dynamically adjusts weights based on changes over time, achieving noise reduction while preserving local instantaneous dynamic changes. Extensive experiments on the CACLE and NASA datasets demonstrate the superiority of the proposed model in terms of RUL. Experimental results show that this method achieves higher prediction accuracy on multiple sets of battery data, while maintaining good robustness and generalization performance in cross-sample tests. Its overall performance is superior to existing comparative models.