Scholarlink

<p>As a core component of modern energy storage systems, lithium-ion batteries play an irreplaceable role in portable electronic devices, new energy vehicles, and renewable energy storage. Precise assessment of the battery’s state of health (SOH) is crucial for ensuring the safety and reliability of system operation. To address this challenge, this paper proposes a novel SOH estimation method that integrates Savitzky-Golay (SG) filter, northern goshawk optimization (NGO), convolutional neural networks (CNN), and bidirectional gated recurrent units (BIGRU). Firstly, the random forest (RF) algorithm ranks and evaluates the significance of representative health features (HFs) extracted from voltage characteristics, current waveforms, and incremental capacity (IC) curves during charging and discharging. This establishes a robust mapping relationship between the HFs and battery capacity degradation. Subsequently, to reduce noise interference and short-term fluctuations caused by capacity regeneration, the SG filter preprocesses the optimized feature subset. This enhances the stability and reliability of input parameters while preserving critical aging-related information. Furthermore, an NGO-CNN-BIGRU framework is proposed for high-precision SOH estimation. NGO optimizes both CNN and BIGRU modules. Specifically, the optimized CNN extracts spatial correlation features from the input data, while the optimized BIGRU captures long-term dependencies in time series through bidirectional information flow. Finally, the effectiveness of the proposed method is validated using the NASA and CACLE datasets, with comparisons to existing methods. The results demonstrate RMSE and MAE values of 0.0072 and 0.0053, respectively, and <i>R</i><sup><i>2</i></sup> exceeding 98.4%. These findings demonstrate significant improvements in accuracy and reliability for SOH estimation.</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

State of health estimation for lithium-ion batteries based on Savitzky Golay filter-NGO-CNN-BIGRU

  • 期刊论文

  • 发表日期：

    2025年8月22日

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

摘要

As a core component of modern energy storage systems, lithium-ion batteries play an irreplaceable role in portable electronic devices, new energy vehicles, and renewable energy storage. Precise assessment of the battery’s state of health (SOH) is crucial for ensuring the safety and reliability of system operation. To address this challenge, this paper proposes a novel SOH estimation method that integrates Savitzky-Golay (SG) filter, northern goshawk optimization (NGO), convolutional neural networks (CNN), and bidirectional gated recurrent units (BIGRU). Firstly, the random forest (RF) algorithm ranks and evaluates the significance of representative health features (HFs) extracted from voltage characteristics, current waveforms, and incremental capacity (IC) curves during charging and discharging. This establishes a robust mapping relationship between the HFs and battery capacity degradation. Subsequently, to reduce noise interference and short-term fluctuations caused by capacity regeneration, the SG filter preprocesses the optimized feature subset. This enhances the stability and reliability of input parameters while preserving critical aging-related information. Furthermore, an NGO-CNN-BIGRU framework is proposed for high-precision SOH estimation. NGO optimizes both CNN and BIGRU modules. Specifically, the optimized CNN extracts spatial correlation features from the input data, while the optimized BIGRU captures long-term dependencies in time series through bidirectional information flow. Finally, the effectiveness of the proposed method is validated using the NASA and CACLE datasets, with comparisons to existing methods. The results demonstrate RMSE and MAE values of 0.0072 and 0.0053, respectively, and R2 exceeding 98.4%. These findings demonstrate significant improvements in accuracy and reliability for SOH estimation.