Scholarlink

To address the variations in voltage characteristics of lithium-ion batteries (LIBs) throughout their aging process and the consequent need for highly accurate remaining useful life (RUL) prediction, this study proposes an innovative prognostics framework that synergistically integrates parameterized modeling with an LSTM neural network. Initially, a state of charge (SOC) normalization scheme under constant-current charge/discharge scenarios is developed to normalize voltage curves onto a common scale. Thereafter, an analytical voltage model is established through polynomial fitting and interpolation techniques, from which an incremental capacity (IC) model is derived to quantitatively capture battery degradation behavior. Moreover, six health indicators (HIs), encompassing IC features and Coulombic efficiency, are extracted as representative metrics to characterize the battery’s performance deterioration. These HIs are then utilized as inputs to an LSTM network for RUL forecasting. Experimental results confirm that the proposed approach outperforms conventional RNN and LSTM models in both capacity estimation and RUL prediction. Specifically, the RMSE for capacity estimation is reduced to 0.0051 Ah, while the RUL prediction error remains within a single cycle, demonstrating superior predictive accuracy and robustness. This work thus presents a practical and effective strategy for advancing the predictive maintenance of LIBs.

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 Remaining Useful Life Prediction Method for Lithium-Ion Batteries by Fusing Parametric Model and LSTM

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Wei Liu,
  • Jiashen Teh,
  • Maoqi Cui,
  • Deyue Meng,
  • Ruixue Shi

摘要

To address the variations in voltage characteristics of lithium-ion batteries (LIBs) throughout their aging process and the consequent need for highly accurate remaining useful life (RUL) prediction, this study proposes an innovative prognostics framework that synergistically integrates parameterized modeling with an LSTM neural network. Initially, a state of charge (SOC) normalization scheme under constant-current charge/discharge scenarios is developed to normalize voltage curves onto a common scale. Thereafter, an analytical voltage model is established through polynomial fitting and interpolation techniques, from which an incremental capacity (IC) model is derived to quantitatively capture battery degradation behavior. Moreover, six health indicators (HIs), encompassing IC features and Coulombic efficiency, are extracted as representative metrics to characterize the battery’s performance deterioration. These HIs are then utilized as inputs to an LSTM network for RUL forecasting. Experimental results confirm that the proposed approach outperforms conventional RNN and LSTM models in both capacity estimation and RUL prediction. Specifically, the RMSE for capacity estimation is reduced to 0.0051 Ah, while the RUL prediction error remains within a single cycle, demonstrating superior predictive accuracy and robustness. This work thus presents a practical and effective strategy for advancing the predictive maintenance of LIBs.