Scholarlink

<p>Accurate prediction of the state of charge (SOC) in lithium-ion batteries is critical for unmanned aerial vehicle (UAV) flight safety and energy management. Current deep learning approaches lack explicit temporal modeling, limiting their ability to predict multi-step SOC trajectories under dynamic condition adaptability. To address these limitations, this paper proposes a hybrid deep learning framework, termed temporal-aware transformer networks (TATNS). The framework combines localized spatio-temporal feature extraction with long-range dependency modeling through temporal encoders and a sliding window-based multi-output mechanism, enhancing prediction accuracy and adaptability in fluctuating environments. Experimental validation was conducted using a large-scale dataset under varying operational conditions. The prediction results demonstrate the superior performance, with mean absolute percentage error values of 4.320, respectively, representing reductions more than 7.97% compared to conventional models including long short-term memory, recurrent neural network, and standalone convolutional neural network/transformer architectures. The model robustness was further verified across diverse temperature scenarios, consistently achieving high prediction accuracy. Compared to traditional deep learning methods, the TATNS framework exhibits enhanced reliability and precision in SOC prediction, demonstrating significant potential for next-generation UAV 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

Temporal-aware transformer networks for state of charge multi-output prediction in unmanned aerial vehicle lithium-ion batteries

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月12日

查看全文
  • Mingzhuang Hua,
  • Miaokun Xu,
  • Jinbo Li,
  • Xiaopeng Zhao,
  • Qin Wang,
  • Xinyuan Ni,
  • Yueqing Wu

摘要

Accurate prediction of the state of charge (SOC) in lithium-ion batteries is critical for unmanned aerial vehicle (UAV) flight safety and energy management. Current deep learning approaches lack explicit temporal modeling, limiting their ability to predict multi-step SOC trajectories under dynamic condition adaptability. To address these limitations, this paper proposes a hybrid deep learning framework, termed temporal-aware transformer networks (TATNS). The framework combines localized spatio-temporal feature extraction with long-range dependency modeling through temporal encoders and a sliding window-based multi-output mechanism, enhancing prediction accuracy and adaptability in fluctuating environments. Experimental validation was conducted using a large-scale dataset under varying operational conditions. The prediction results demonstrate the superior performance, with mean absolute percentage error values of 4.320, respectively, representing reductions more than 7.97% compared to conventional models including long short-term memory, recurrent neural network, and standalone convolutional neural network/transformer architectures. The model robustness was further verified across diverse temperature scenarios, consistently achieving high prediction accuracy. Compared to traditional deep learning methods, the TATNS framework exhibits enhanced reliability and precision in SOC prediction, demonstrating significant potential for next-generation UAV systems.