A real-time mapping method for knowledge graph-driven large language models: a focus on indoor fire evacuations

Efficient indoor fire evacuations are crucial for protecting human life and property. Traditional indoor fire evacuation methods often do not fully account for individual scenario differences, dynamic environments, and their impact on evacuation behavior, leading to evacuation routes that lack flexibility and real-time responsiveness. This study introduces a real-time mapping method driven by a knowledge graph and large language model. The method combines an indoor fire knowledge graph with ant colony algorithms, enabling real-time perception of fire dynamics to generate personalized evacuation route maps and guidance. A prototype system is developed and subjected to experimental analysis. The experimental results show that this system significantly outperforms the standard LLM in handling indoor fire-related information. Moreover, in evacuation simulation experiments, participants using this system experienced a 27.41% increase in evacuation efficiency, and a significant decrease in anxiety levels. This approach effectively enhances the public's ability to respond to emergencies during indoor fires. Additionally, this study offers new insights for other complex dynamic scenarios that require decision support.