Real-time prediction of tunnel fire physical fields based on deep Learning: Integrating ceiling smoke extraction scenarios

Tunnels are critical transportation infrastructures. In the event of a tunnel fire, rapidly formulating an optimal smoke extraction strategy is an urgent challenge. This paper focuses on fast prediction methods for smoke propagation in tunnel fire scenarios, particularly under ceiling smoke extraction conditions, to enhance the response speed of emergency evacuation and rescue operations. To tackle the scarcity of tunnel fire-related data, we construct the Tunnel Fire Dataset (TFD) using numerical simulation techniques. Our results demonstrate that deep learning can accurately predict the dynamic evolution of multiple physical fields (such as temperature, visibility, and velocity) during tunnel fire incidents. This capability is crucial for optimizing smoke extraction systems and formulating effective fire response strategies. By leveraging deep learning models, we achieve a structural similarity index (SSIM) exceeding 0.99 for key physical field predictions, maintain mean squared error (MSE) within a small range across different data segments, and improve single-step inference speed to the second level. These advancements significantly enhance the timeliness and accuracy of tunnel fire emergency response. This research provides an improved pathway for emergency management and safety planning in complex fire scenarios.