Scholarlink

<p>Tunnel fires pose severe threats to life safety due to the rapid spread of toxic, high-temperature smoke. Effective ventilation control is critical for safe evacuation, yet designing optimal strategies requires a precise understanding of complex smoke behavior. This paper develops a hybrid deep learning model, integrating Convolutional and Long Short-Term Memory (CNN-LSTM) networks, to achieve accurate prediction of key smoke properties including temperature, visibility, and smoke layer height under varied ventilation conditions. The model was trained on a comprehensive database of 204 fire scenarios generated through high-resolution CFD simulations. Small-scale tunnel fire tests validated the model’s performance, with results showing excellent agreement with experimental data. The model demonstrated high overall accuracy, with particularly outstanding performance for temperature and smoke layer height predictions, while visibility predictions remained at an acceptable level, consistently capturing overall trends. This AI-driven model demonstrates the potential for integration into future intelligent ventilation control systems, enabling dynamic smoke management and providing rapid, reliable predictions to support decision-making during tunnel fire emergencies.</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

Spatio-Temporal Prediction of Critical Smoke Properties in Tunnel Fires Using a Hybrid CNN-LSTM Network

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Shiqiang Deng,
  • Hao Ding,
  • Maogui Sun,
  • Chao Hu,
  • Shuai Liu,
  • Meng Yang,
  • Shuping Jiang

摘要

Tunnel fires pose severe threats to life safety due to the rapid spread of toxic, high-temperature smoke. Effective ventilation control is critical for safe evacuation, yet designing optimal strategies requires a precise understanding of complex smoke behavior. This paper develops a hybrid deep learning model, integrating Convolutional and Long Short-Term Memory (CNN-LSTM) networks, to achieve accurate prediction of key smoke properties including temperature, visibility, and smoke layer height under varied ventilation conditions. The model was trained on a comprehensive database of 204 fire scenarios generated through high-resolution CFD simulations. Small-scale tunnel fire tests validated the model’s performance, with results showing excellent agreement with experimental data. The model demonstrated high overall accuracy, with particularly outstanding performance for temperature and smoke layer height predictions, while visibility predictions remained at an acceptable level, consistently capturing overall trends. This AI-driven model demonstrates the potential for integration into future intelligent ventilation control systems, enabling dynamic smoke management and providing rapid, reliable predictions to support decision-making during tunnel fire emergencies.