Abstract: In building fires, the rapid spread of smoke is one of the major factors leading to casualties and difficulties in rescue operations. This paper presents a rapid fire smoke field modeling method based on smoke sensor data. By acquiring the real-time optical extinction coefficient data from sensors, combined with a dual-agent deep learning model, we predict the fire source location (R2=97%) and smoke density field (R2=91%), and dynamically reconstruct the 3D smoke field using volume rendering technology in the virtual reality engine, Unreal Engine. This method significantly reduces the manpower and computational resource burden required for constructing smoke field scenes in traditional virtual reality systems, overcoming the technical bottlenecks of low modeling efficiency, complex operation processes, and lack of real-time control in conventional methods. It achieves rapid generation and dynamic visualization updates of the smoke field. This method is not only applicable for virtual fire-fighting training and behavioral simulation but also has excellent scalability, making it suitable for integration into smart building fire response systems and city-level digital twin platforms.

Key words: fire smoke field modeling, real-time visualization, fire source localization, deep learning, volume rendering