Scholarlink

<p>Remaining useful life (RUL) prediction of lithium-ion batteries is a critical task in battery management systems. To overcome the limitations of existing methods in feature construction, prediction accuracy, and early-stage small-sample learning, an RUL prediction approach integrating a long short-term memory (LSTM) network with a dynamic forking lightning search algorithm (DFLSA) is proposed. Temperature-voltage coupled features, including differential thermal voltammetry (DTV)-based features, are constructed based on battery degradation mechanisms, and key features highly correlated with capacity are selected using the Pearson correlation coefficient. DFLSA, incorporating a dynamic forking probability adjustment strategy, is employed to optimize the hyperparameters of the LSTM model, thereby enhancing prediction performance. Experimental results on both the Oxford and NASA battery datasets demonstrate that the proposed DFLSA-LSTM model achieves high prediction accuracy and robustness, indicating strong adaptability and practical value for engineering applications.</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

Early-stage RUL prediction of lithium-ion batteries using DFLSA-enhanced LSTM with temperature–voltage coupled features

  • 期刊论文

  • 发表日期：

    2026年4月29日

查看全文
  • Yao Zhao,
  • Yuhong Jiang,
  • Zhaoying He,
  • Wenyi Liu

摘要

Remaining useful life (RUL) prediction of lithium-ion batteries is a critical task in battery management systems. To overcome the limitations of existing methods in feature construction, prediction accuracy, and early-stage small-sample learning, an RUL prediction approach integrating a long short-term memory (LSTM) network with a dynamic forking lightning search algorithm (DFLSA) is proposed. Temperature-voltage coupled features, including differential thermal voltammetry (DTV)-based features, are constructed based on battery degradation mechanisms, and key features highly correlated with capacity are selected using the Pearson correlation coefficient. DFLSA, incorporating a dynamic forking probability adjustment strategy, is employed to optimize the hyperparameters of the LSTM model, thereby enhancing prediction performance. Experimental results on both the Oxford and NASA battery datasets demonstrate that the proposed DFLSA-LSTM model achieves high prediction accuracy and robustness, indicating strong adaptability and practical value for engineering applications.