IDSS-Driven Quantitative Risk Assessment and Dynamic Evacuation Routing for Train Fires in Railway Bridge–Tunnel Connection Sections

Train fires in railway bridge–tunnel connection sections (BTCSs) create severe evacuation challenges because tunnel–bridge spatial transitions interact with heat, smoke, visibility loss, and constrained rescue conditions. Existing evacuation management methods remain limited in coupling quantitative risk assessment with adaptive route guidance under evolving fire hazards. To address this issue, this paper proposes a large language model (LLM)-enhanced intelligent decision-support system (IDSS) framework for quantitative risk assessment and dynamic evacuation routing in BTCS fire scenarios. First, a multi-dimensional risk assessment model is established using the analytic hierarchy process and fuzzy comprehensive evaluation to quantify post-stop evacuation risk from the perspectives of evacuation organization, structural damage, and line recovery. Second, a dynamic topology-based routing method is developed to prune fire-threatened nodes and identify safer evacuation paths under evolving hazard conditions. The risk assessment model and routing algorithm are further embedded as callable tools into an LLM-enhanced evacuation IDSS under a perception–reasoning–recommendation architecture, in which an LLM orchestrates tool invocation, situational reasoning, and recommendation generation, thereby enabling autonomous risk interpretation, dynamic route replanning, and cross-regional collaborative decision support. The proposed framework is validated through a representative real-world railway engineering case. The results show that the IDSS-recommended routes achieved higher comprehensive safety scores (80.44 and 79.56) than routes involving fire-affected areas did (77.00 and 77.88). Workflow analysis further indicates that the proposed IDSS reduces the manual route-derivation workload by integrating risk assessment, topology pruning, and route allocation into structured, human-reviewable evacuation recommendations. Expert evaluations further confirm the rationality and compliance of the outputs, with review scores ranging from 1.76 to 1.92 out of 2.00. Overall, the proposed framework offers a feasible decision-support approach for intelligent evacuation management in complex railway fire emergencies.