Abstract
<title>Abstract</title> <p> The flight deck of a naval vessel is a high-risk operational environment in which a single initiating event can precipitate coupled hazard cascades encompassing concurrent fire propagation, progressive casualty deterioration, and damage-induced stability degradation. Conventional pre-plan-driven response approaches are architecturally ill-suited to such multi-hazard, temporally evolving scenarios, as they cannot adapt to stochastic hazard interactions in real time. This paper presents the Intelligent Emergency Decision-support Framework for Flight-deck Safety (IEDFS), a closed-loop architecture comprising three tightly coupled modules: (1) a knowledge–data dual-driven dynamic reasoning engine that performs online fire propagation modelling, casualty condition state estimation, real-time confidence updating, and damage stability constraint evaluation; (2) an attention-enhanced multi-agent cooperative decision module in which the task-allocation problem is formalised as a Decentralised Partially Observable Markov Decision Process (Dec-POMDP) and solved via an Attention-QMIX network preserving heterogeneous agent role attributes; and (3) a hierarchical time-varying path planning module combining a temporally extended A* algorithm for global route generation with D* Lite for local dynamic replanning, enabling simultaneous hazard avoidance, stretcher-carry task scheduling, and inter-agent conflict resolution. At each decision cycle, the reasoning engine furnishes updated situational assessments to the decision module, whose cooperative policy outputs are consumed by the path planner—completing a feedback loop that sustains adaptive response under continuously evolving hazard states. Validation on the FDAS-Sim platform across five progressively escalating accident scenarios demonstrates that <italic>IEDFS improves the critical-casualty golden-hour treatment rate by</italic> <bold>26.3 percentage points</bold> , <italic>reduces fire containment time by</italic> <bold>40.1%</bold> , <italic>decreases flight deck operational recovery time by</italic> <bold>36.1%</bold> , <italic>and lowers the damage stability exceedance rate by</italic> <bold>53.2%</bold> <italic>relative to static rule-based baselines.</italic> Ablation studies confirm the independent contribution of each module and quantify the synergistic performance gain from system integration. These results position IEDFS as a viable methodological foundation for intelligent decision support and simulation-based proficiency training in flight deck emergency response. </p>