Scholarlink

<p>Accurate state of charge (SOC) estimation is essential for the safe and efficient operation of lithium-ion battery management systems. However, maintaining both accuracy and robustness under varying temperatures and complex operating conditions remains challenging. To address this, the study proposes a GF-TCN-Transformer model that integrates Gaussian filtering (GF) with a hybrid neural network. Gaussian filtering is employed for data augmentation to learn slow-varying information in voltage and current signals. A temporal convolutional network (TCN) is employed to capture local temporal dependencies, while a Transformer models long-range correlations, enabling high-precision SOC prediction. To evaluate generalization, we adopt dynamic stress test (DST) and Beijing bus dynamic stress test (BBDST) experiments under multiple temperatures and further introduce public datasets from Panasonic and the University of Maryland. The proposed model achieves maximum SOC errors of 1.90%, 1.18%, 2.20%, and 1.51% across different temperatures. On public datasets with complex operating profiles, the maximum error remains within 1.61%. Overall, the GF-TCN-Transformer demonstrates superior accuracy and robustness across diverse thermal and dynamic conditions, offering a reliable solution for SOC estimation and supporting the development of intelligent battery management systems.</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 Gaussian filtering enhanced TCN-Transformer hybrid network for robust state of charge estimation of lithium-ion batteries under complex operating conditions

  • 期刊论文

  • 发表日期：

    2026年7月8日

查看全文
  • Qiao Liu,
  • Zhuo Zhang,
  • Xinyue Shu,
  • Mingyang Zhang,
  • Xuelin Xu

摘要

Accurate state of charge (SOC) estimation is essential for the safe and efficient operation of lithium-ion battery management systems. However, maintaining both accuracy and robustness under varying temperatures and complex operating conditions remains challenging. To address this, the study proposes a GF-TCN-Transformer model that integrates Gaussian filtering (GF) with a hybrid neural network. Gaussian filtering is employed for data augmentation to learn slow-varying information in voltage and current signals. A temporal convolutional network (TCN) is employed to capture local temporal dependencies, while a Transformer models long-range correlations, enabling high-precision SOC prediction. To evaluate generalization, we adopt dynamic stress test (DST) and Beijing bus dynamic stress test (BBDST) experiments under multiple temperatures and further introduce public datasets from Panasonic and the University of Maryland. The proposed model achieves maximum SOC errors of 1.90%, 1.18%, 2.20%, and 1.51% across different temperatures. On public datasets with complex operating profiles, the maximum error remains within 1.61%. Overall, the GF-TCN-Transformer demonstrates superior accuracy and robustness across diverse thermal and dynamic conditions, offering a reliable solution for SOC estimation and supporting the development of intelligent battery management systems.