Scholarlink

<p>Accurately estimating the remaining useful life (RUL) of a battery is important to ensure the safety and reliability of battery system operation. This paper presents a novel framework to improve the RUL prediction accuracy of lithium-ion batteries (LIBs) by integrating the multi-head attention (MHA) mechanism into a bidirectional long-short-term memory (BILSTM) network to form a parallel BILSTM-MHA neural network. Firstly, the proposed RUL prediction framework establishes various features over the life cycle of LIBs, which are initially used as inputs to a multi-input, single-output BILSTM-MHA model for capacity estimation. Subsequently, the parallel BILSTM-MHA neural network model based on the&#xa0;adaptive noise-based&#xa0;empirical mode decomposition (ANEMD) module utilizes the estimated capacity data to perform the RUL prediction of LIBs. The described prediction framework maximizes the utilization of diverse features of LIBs over their life cycle and optimizes the information extraction from the data obtained from complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) decomposition through a parallel network. This network considers the capacity regeneration effect, leading to a significant improvement in LIBs' RUL prediction accuracy. Two case studies are performed using the NASA battery dataset and&#xa0;the CALCE battery dataset,&#xa0;which demonstrate that the proposed method outperforms traditional deep learning models in predicting&#xa0;the RUL of LIBs.</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 batteries remaining useful life prediction using a parallel BILSTM-MHA neural network based on a CEEMDAN module

  • 期刊论文

  • 发表日期：

    2025年3月13日

查看全文
  • Chaoqun Duan,
  • Hengrui Cao,
  • Fuqiang Liu,
  • Xin Li,
  • Xuelian Duan,
  • Bo Sheng

摘要

Accurately estimating the remaining useful life (RUL) of a battery is important to ensure the safety and reliability of battery system operation. This paper presents a novel framework to improve the RUL prediction accuracy of lithium-ion batteries (LIBs) by integrating the multi-head attention (MHA) mechanism into a bidirectional long-short-term memory (BILSTM) network to form a parallel BILSTM-MHA neural network. Firstly, the proposed RUL prediction framework establishes various features over the life cycle of LIBs, which are initially used as inputs to a multi-input, single-output BILSTM-MHA model for capacity estimation. Subsequently, the parallel BILSTM-MHA neural network model based on the adaptive noise-based empirical mode decomposition (ANEMD) module utilizes the estimated capacity data to perform the RUL prediction of LIBs. The described prediction framework maximizes the utilization of diverse features of LIBs over their life cycle and optimizes the information extraction from the data obtained from complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) decomposition through a parallel network. This network considers the capacity regeneration effect, leading to a significant improvement in LIBs' RUL prediction accuracy. Two case studies are performed using the NASA battery dataset and the CALCE battery dataset, which demonstrate that the proposed method outperforms traditional deep learning models in predicting the RUL of LIBs.