Study Numerical Simulation of Smoke Movement and LSTM-Based Early Warning for metro Platform Fires
This study integrates Fire Dynamics Simulator (FDS) numerical simulations, full-scale smoke tests, and a Long Short-Term Memory (LSTM) deep learning model to analyze smoke dispersion in metro platform fires and develop a highly accurate early warning system that achieves over 98% accuracy in predicting critical visibility drops.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Underground cities are a marvel of modern engineering, moving millions of people daily through tunnels that lie deep beneath the surface. Yet, this convenience comes with a hidden vulnerability: when a fire breaks out in a subway station, the smoke it produces becomes the primary threat to life. Unlike a fire in an open field where heat and fumes can rise and dissipate, a subway station is a confined box. Without natural wind to push the smoke away, it fills the space rapidly, obscuring vision and filling the air with toxic gases. The speed at which smoke spreads determines whether passengers can find their way to safety or become trapped. For decades, engineers have relied on complex ventilation systems to manage this danger, but the physics of smoke in these tight, crowded spaces is difficult to predict, and the window for a safe escape is often measured in mere minutes.
To address this challenge, a team of researchers from China and Canada recently conducted a comprehensive study to understand exactly how smoke behaves during a platform fire and how to warn people before it is too late. They combined three distinct approaches: building a detailed digital twin of a subway station, performing a real-life fire test in an operating station, and training a computer to recognize danger patterns. Their goal was not just to see how smoke moves, but to develop a system that could predict a hazardous drop in visibility seconds before it happens, giving emergency responders and passengers a crucial head start.
The researchers began by creating a precise computer model of a typical double-level subway station, similar to those found in many major Chinese cities. They simulated a fire on the platform with a heat output of 3 megawatts, a size comparable to a large vehicle fire, and watched how the smoke spread under different conditions. In one scenario, they used the station's standard ventilation system. In another, they added a mobile exhaust fan to see if it could help clear the air faster. In a third scenario, they modeled a station without the glass barriers that separate the platform from the tracks, a design common in older systems in cities like New York and London. The simulations revealed that the presence of these glass doors, known as platform screen doors, fundamentally changes how smoke travels. When the doors are closed, smoke is trapped on the platform until the system opens them to let the air flow into the tunnel. When the doors are absent, smoke spreads immediately into the tunnel and up toward the surface, creating a different, often more chaotic, pattern of movement.
To ensure their computer models were accurate, the team did not rely on theory alone. They conducted a full-scale fire experiment in a real, operating subway station in southern China during the night when trains were not running. They ignited a controlled fire using four trays of burning alcohol, generating a heat release rate of 1.5 megawatts. As the fire grew, they watched the smoke rise and spread, recording the exact moment the ventilation systems kicked in. The experiment confirmed that the station's safety logic worked as designed: the smoke detectors triggered the exhaust fans, the glass doors opened to connect the platform to the tunnel, and the tunnel's own ventilation system pulled the smoke away. The smoke was successfully guided from the platform, through the open doors, and out of the station, leaving the platform clear enough for evacuation within minutes. This real-world test proved that their digital simulations were reliable enough to be used for further analysis.
With a validated model and a wealth of data, the researchers turned to the future of fire safety: artificial intelligence. They fed thousands of data points from their simulations into a specialized type of computer program called a Long Short-Term Memory network, or LSTM. This program is designed to learn from sequences of events, much like how a human learns to recognize a pattern over time. They trained the system to look at the levels of carbon monoxide, carbon dioxide, and visibility at various points on the platform. The goal was for the computer to predict if the visibility would drop below a critical safety threshold of 10 meters within the next 30 seconds.
The results were striking. When tested on new data it had never seen before, the computer model correctly identified dangerous conditions with an accuracy of over 98 percent. More importantly, it rarely missed a real danger. Out of 118 simulated fire incidents where visibility dropped to unsafe levels, the model triggered an alarm for every single one. In about 34 percent of these cases, the system sounded the warning before the visibility actually reached the critical point, providing an average advance notice of nearly 26 seconds. While this may seem like a short time, in the context of a crowded subway platform where panic can spread quickly, those seconds are vital. They are enough time to activate emergency broadcasts, switch ventilation modes, and begin guiding passengers toward safety before the air becomes too thick to see through.
The study also highlighted the value of combining different tools. The real-world fire test validated the computer models, and the computer models provided the massive dataset needed to train the artificial intelligence. The researchers noted that while their current model is highly effective, it is a single-layer system designed for speed. Future work will involve adding more data from different types of fires and potentially combining this system with other advanced techniques to make it even more robust. For now, the study offers a clear path forward: by understanding the physics of smoke through simulation and testing, and then using that knowledge to train intelligent warning systems, cities can make their underground transit networks significantly safer for the millions who rely on them every day.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.