Scholarlink

<p>It is critical to accurately predict the capacity and remaining useful life (RUL) of lithium-ion batteries (LIBs) for reliable operation and timely maintenance of electric vehicles. However, challenges persist due to the uncertainty in battery capacity degradation trends and interference from external noise. This study suggests a novel neural network model, DPformer, to capture the capacity fade trend and reduce interference from external noise, which integrates feature reconstruction, attention mechanism, and combined multi-layer perception. Firstly, the raw data is reconstructed and denoised by an automatic denoising encoder (ADE), and the long-term dependencies in time series information are effectively captured via the attention mechanism. Subsequently, the extracted multi-scale information is further processed by a designed feature pyramid decoder (FPD) to achieve better feather representations. In addition, a particle swarm optimization algorithm is improved to optimize the hyperparameters of the proposed model more precisely. Finally, the performance of the proposed is validated by using two public datasets. Experimental results demonstrate that the model achieves good performances in prediction accuracy and generalizability, and achieves up to 30–50% improvement in terms of relative error (RE).</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 novel remaining useful life prediction for the lithium-ion battery using DPformer and enhanced optimization techniques

  • 期刊论文

  • 发表日期：

    2025年2月19日

查看全文
  • Delin Huang,
  • Qiuyu Ran,
  • Jinghui Yang,
  • Dexian Wang,
  • Xiangdong Su

摘要

It is critical to accurately predict the capacity and remaining useful life (RUL) of lithium-ion batteries (LIBs) for reliable operation and timely maintenance of electric vehicles. However, challenges persist due to the uncertainty in battery capacity degradation trends and interference from external noise. This study suggests a novel neural network model, DPformer, to capture the capacity fade trend and reduce interference from external noise, which integrates feature reconstruction, attention mechanism, and combined multi-layer perception. Firstly, the raw data is reconstructed and denoised by an automatic denoising encoder (ADE), and the long-term dependencies in time series information are effectively captured via the attention mechanism. Subsequently, the extracted multi-scale information is further processed by a designed feature pyramid decoder (FPD) to achieve better feather representations. In addition, a particle swarm optimization algorithm is improved to optimize the hyperparameters of the proposed model more precisely. Finally, the performance of the proposed is validated by using two public datasets. Experimental results demonstrate that the model achieves good performances in prediction accuracy and generalizability, and achieves up to 30–50% improvement in terms of relative error (RE).