Abstract: To consider the dynamic characteristics of underground fire occurrence and development, combining with the changes of temperature and CO volume fraction over time, a real-time evacuation path planning method for underground space was established based on A* algorithm, which considered the evolution of fire scenarios. The path finding results of Pathfinder software, ASTAR algorithm and ant colony algorithm for both a virtual small-scale environment and a full-scale underground metro station were compared, from which the efficiency and effectiveness of A* algorithm were verified. By establishing the time-history database of some key indices, the possible regions for evacuation was identified based on the corresponding tolerance limit of human being. The A* algorithm was then adopted to find the optimal evacuation path based on the reconstructed map. Thereby, the real-time evacuation path planning for underground space during a fire accident was accomplished. The proposed approach was used for the dynamic evacuation path planning of a deep underground metro station. The real-time data of temperature and CO concentration were derived from fire accident simulations. The optimal evacuation path and regions where evacuation was urgent were identified, and the maximum evacuation time was determined. The proposed approach overcomes the limitation of static path planning and can provide important guidance for the real-time path planning and emergency strategies of deep underground space subject to fire accident.