Scholarlink

<p>The precise prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for optimizing operational efficiency and safeguarding equipment safety. This paper presents a novel approach for predicting the RUL of lithium-ion batteries through advanced feature engineering and a hybrid predictive model. Firstly, to acquire high-quality health features (HFs), a diverse array of representative HFs is derived from charge and discharge curves, with specific extraction of those related to peak values from the IC curves. The random forest (RF) algorithm is utilized to rank and assess the significance of the HFs. Secondly, to tackle the challenges of noise and capacity regeneration in the original data, HFs are decomposed and reconstructed utilizing the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN). Thirdly, to achieve accurate RUL predictions, a hybrid model is proposed integrating convolutional neural networks (CNN), bi-directional gated recurrent units (BiGRU), and attention mechanisms (AM). The CNN-BiGRU-AM model integrates spatiotemporal feature learning with adaptive attention weighting, thereby effectively avoiding local optima and enhancing the accuracy of lithium-ion battery RUL prediction and generalization performance. Finally, the effectiveness of the proposed approach is substantiated through a comparative analysis with several other methods utilizing the CALCE dataset. The results demonstrate that the model’s RMSE, MAE, and MAPE are maintained at 0.0043, 0.0028, and 0.0018, respectively, indicating significant improvements in RUL predictive accuracy with high reliability.</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

Research on the remaining useful life prediction method for lithium-ion batteries based on feature engineering and CNN-BiGRU-AM model

  • 期刊论文

  • 发表日期：

    2025年4月24日

查看全文
  • Di Zheng,
  • Ye Zhang,
  • Xifeng Guo,
  • Yi Ning,
  • Rongjian Wei

摘要

The precise prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for optimizing operational efficiency and safeguarding equipment safety. This paper presents a novel approach for predicting the RUL of lithium-ion batteries through advanced feature engineering and a hybrid predictive model. Firstly, to acquire high-quality health features (HFs), a diverse array of representative HFs is derived from charge and discharge curves, with specific extraction of those related to peak values from the IC curves. The random forest (RF) algorithm is utilized to rank and assess the significance of the HFs. Secondly, to tackle the challenges of noise and capacity regeneration in the original data, HFs are decomposed and reconstructed utilizing the improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN). Thirdly, to achieve accurate RUL predictions, a hybrid model is proposed integrating convolutional neural networks (CNN), bi-directional gated recurrent units (BiGRU), and attention mechanisms (AM). The CNN-BiGRU-AM model integrates spatiotemporal feature learning with adaptive attention weighting, thereby effectively avoiding local optima and enhancing the accuracy of lithium-ion battery RUL prediction and generalization performance. Finally, the effectiveness of the proposed approach is substantiated through a comparative analysis with several other methods utilizing the CALCE dataset. The results demonstrate that the model’s RMSE, MAE, and MAPE are maintained at 0.0043, 0.0028, and 0.0018, respectively, indicating significant improvements in RUL predictive accuracy with high reliability.