Abstract: As an authoritative standard for design, construction, and supervision, fire protection standards play a crucial role in ensuring work safety. However, the interpretation and implementation of standards have relied on the experienced operations of professionals and manual verification. To address this issue, this paper designs a question-answering system for fire protection standards based on large language models and retrieval augmented generation, realizing the knowledge question-answering function in the professional field of fire protection standards and specifications. The system consists of a large language model, an external knowledge base, and a user interface. To improve the retrieval quality of the knowledge base, a retrieval algorithm weighted by professional terms is proposed. A set of question optimization templates suitable for fire protection standards is constructed to enhance the accuracy of users' question and the quality of the system's answers. Considering data security, a lightweight local deployment is achieved. The practical tests show that the question-answering system can accurately answer questions related to fire protection standards and has good stability and practicality. The development of system has opened up new ideas for the intelligent expression recognition and content analysis of fire protection standards, and it represents a beneficial exploration of large language model technology in the field of smart fire protection.

Key words: fire protection standards, question and answer system, large language model, retrieval augmented generation, smart fire protection