Scholarlink

<p>Lithium-ion batteries, as critical components of spacecraft attitude control systems, must meet stringent safety and reliability requirements during repeated mission reentries. To address the challenges of limited data samples and insufficient prior physical knowledge in conventional remaining useful life (RUL) prediction approaches-particularly the performance degradation of data-driven methods caused by the small number of flight missions-this study proposes a hybrid model–data-driven framework integrating a Gamma-process–based degradation model with an improved Transformer network. The key novelty lies in coupling model-driven RUL point estimation and probabilistic prior information with a sequence-to-sequence Transformer prediction framework under a unified multi-source health-factor representation. First, multiple health indicators are constructed from discharge curves and temperature rise profiles. Subsequently, a stochastic degradation model based on the Gamma process is formulated using these indicators, and an empirical maximum likelihood algorithm combined with particle filtering is employed to estimate the parameters of the remaining life distribution. Finally, a sequence-to-sequence direct mapping method is developed by enhancing the Transformer architecture to infer RUL values from measurable data. Experiments conducted on a publicly available battery dataset demonstrate that the proposed method achieves superior RUL prediction accuracy compared with existing approaches.</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

A Hybrid Model–Data-Driven Method Integrating Gamma Process and Transformer for Remaining Useful Life Prediction of Lithium-Ion Batteries

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Chaoyue Zhao,
  • Shaojie Ai,
  • Yaxuan Fu,
  • Jia Song

摘要

Lithium-ion batteries, as critical components of spacecraft attitude control systems, must meet stringent safety and reliability requirements during repeated mission reentries. To address the challenges of limited data samples and insufficient prior physical knowledge in conventional remaining useful life (RUL) prediction approaches-particularly the performance degradation of data-driven methods caused by the small number of flight missions-this study proposes a hybrid model–data-driven framework integrating a Gamma-process–based degradation model with an improved Transformer network. The key novelty lies in coupling model-driven RUL point estimation and probabilistic prior information with a sequence-to-sequence Transformer prediction framework under a unified multi-source health-factor representation. First, multiple health indicators are constructed from discharge curves and temperature rise profiles. Subsequently, a stochastic degradation model based on the Gamma process is formulated using these indicators, and an empirical maximum likelihood algorithm combined with particle filtering is employed to estimate the parameters of the remaining life distribution. Finally, a sequence-to-sequence direct mapping method is developed by enhancing the Transformer architecture to infer RUL values from measurable data. Experiments conducted on a publicly available battery dataset demonstrate that the proposed method achieves superior RUL prediction accuracy compared with existing approaches.