Abstract: To address the problems of simple data query mode with insufficient flexibility and poor user-friendliness in the railway natural disaster monitoring system, this paper optimizes its query function by adopting multi-agent technology based on Large Language Model (LLM).Three types of core business data query agents are designed for site data, sensor data and early warning data. A centralized architecture is adopted to form a multi-agent network, which realizes natural language-based data query for the railway natural disaster monitoring system and provides users with a new human-computer interaction mode for data query.On the premise of ensuring the stable operation of the original monitoring system architecture and core functions, this new data query function effectively simplifies the operation process of data query in the system, and significantly improves the flexibility of the query function and the friendliness of result presentation, which can serve as a practical reference for the application of LLM and multi-agent technology in railway information systems.