Scholarlink

<p>Thermal runaway in lithium-ion batteries poses severe safety risks, making early detection and prevention essential for both grid-scale energy storage and electric vehicle applications. This study proposes a Transformer-based prediction method for early warning of lithium-ion battery thermal runaway. The proposed detection method is a Transformer-based prediction architecture that uses the Wild Horse Optimizer (WHO) for parameter optimization. The developed method predicts the onset of battery failure using multi-sensor measurements acquired under abusive operating conditions. A cyclic overcharge experiment was conducted on lithium iron phosphate batteries, during which voltage, temperature, and macrostrain signals were continuously recorded to capture the electrochemical, thermal, and mechanical response preceding thermal runaway. The proposed prediction framework employs a Transformer-based neural network, with its key parameters tuned using a metaheuristic optimization strategy to improve convergence and generalization. Its performance is benchmarked against established data-driven models, including Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Backpropagation Neural Network (BPNN). The results demonstrate that the proposed approach achieves superior accuracy across mean square error, root mean square error, mean absolute error, and coefficient of determination, while maintaining robust performance under normal operating conditions. Notably, the system provides a reliable early warning of imminent failure with a prediction horizon of approximately 13&#xa0;s before thermal runaway during overcharge. The “13 seconds” refers to the time elapsed after the onset of overcharging under the cyclic overcharge experimental condition. These findings show that combining multi-sensor monitoring with advanced data-driven analysis can significantly improve the early detection of hazardous conditions in lithium-ion batteries. The proposed framework offers a practical basis for enhancing protection strategies and safety design in lithium-ion battery energy storage and electric mobility applications.</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

Multi-signal predictive approach for thermal runaway prevention in lithium-ion batteries

  • 期刊论文

  • 发表日期：

    2026年5月6日

查看全文
  • Ganxing Zhang,
  • Lei Han,
  • Meng Ruizhi,
  • Laiqing Yan,
  • Zia Ullah

摘要

Thermal runaway in lithium-ion batteries poses severe safety risks, making early detection and prevention essential for both grid-scale energy storage and electric vehicle applications. This study proposes a Transformer-based prediction method for early warning of lithium-ion battery thermal runaway. The proposed detection method is a Transformer-based prediction architecture that uses the Wild Horse Optimizer (WHO) for parameter optimization. The developed method predicts the onset of battery failure using multi-sensor measurements acquired under abusive operating conditions. A cyclic overcharge experiment was conducted on lithium iron phosphate batteries, during which voltage, temperature, and macrostrain signals were continuously recorded to capture the electrochemical, thermal, and mechanical response preceding thermal runaway. The proposed prediction framework employs a Transformer-based neural network, with its key parameters tuned using a metaheuristic optimization strategy to improve convergence and generalization. Its performance is benchmarked against established data-driven models, including Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Backpropagation Neural Network (BPNN). The results demonstrate that the proposed approach achieves superior accuracy across mean square error, root mean square error, mean absolute error, and coefficient of determination, while maintaining robust performance under normal operating conditions. Notably, the system provides a reliable early warning of imminent failure with a prediction horizon of approximately 13 s before thermal runaway during overcharge. The “13 seconds” refers to the time elapsed after the onset of overcharging under the cyclic overcharge experimental condition. These findings show that combining multi-sensor monitoring with advanced data-driven analysis can significantly improve the early detection of hazardous conditions in lithium-ion batteries. The proposed framework offers a practical basis for enhancing protection strategies and safety design in lithium-ion battery energy storage and electric mobility applications.