HyFiD: LLM-ML Hybrid Framework for Subway Fire Detection

Fire detection is vital in subway tunnels where confined geometries and ventilation complicate safety monitoring. Existing methods, such as classical machine learning, detect fire based on multivariate correlations but often lack the contextual reasoning required for disambiguation. This limitation becomes critical when HVAC-driven airflow disrupts thermal stratification and dilutes gas concentrations, creating ambiguous patterns that mimic fire signatures. Recent studies further suggest that Large Language Models (LLMs) can overcome this bottleneck by leveraging pretrained knowledge to interpret complex sensor dynamics into concise, semantic descriptions. To effectively integrate this semantic capability into a robust detection framework, we propose HyFiD, a hybrid framework that employs an LLM as a semantic feature extractor to augment classical classifiers. By converting momentary multi-sensor readings (temperature, smoke, O2, CO, and CO2) into textual assessments of environmental states, HyFiD generates semantic vectors that are fused with numerical features for robust training. Experiments on Fire Dynamics Simulator (FDS)-based subway scenarios, including ventilation-dominated HVAC events and high-energy battery fires, reveal that semantic augmentation improves performance over numerical-only baselines and prompt-only LLM classifiers, while enhancing interpretability through explicit, human-readable evidence.