土木工程与管理学报2026，Vol.43Issue(1)：108-117,10.DOI:10.13579/j.cnki.2095-0985.2026.20250142
CE-ISMLLM:用于知识检索与事故预防的土木工程智能安全管理大语言模型框架
CE-ISMLLM:Civil Engineering Intelligent Safety Management Large Language Model Framework for Knowledge Retrieval and Accident Prevention
摘要
Abstract
With the rapid development of generative artificial intelligence in the era of Industry 4.0,large language model(LLM)has been widely applied to various vertical domains.To build a proprie-tary RAG-based LLM for intelligent safety management in civil engineering,the primary priority is to segment and embed expert knowledge text of civil engineering management into the basic LLM's vec-tor database without pre-training or fine-tuning.By combining prompt enhancement and expert evalua-tion mechanisms,the output of the LLM undergoes iterative refinement through generative adversarial iterations,ultimately yielding an ideal response.The performance of the LLM is tested by specific pro-fessional questions and real accident cases.Various types of questions and accident details are used as query inputs.The responses generated by the LLM are quantitatively evaluated through multiple indi-cators.Test results indicate that the proposed civil engineering intelligent safety management LLM(CE-ISMLLM)framework has more advantages in knowledge retrieval and accident prevention than general LLM.关键词
大语言模型/检索增强生成/知识检索/事故预防/土木工程智能安全管理Key words
large language model/retrieval-augmented generation/knowledge retrieval/accident prevention/civil engineering intelligent safety management分类
建筑与水利引用本文复制引用
朱文锐,于军琪,翁彤彤,宋正伟,董芳楠..CE-ISMLLM:用于知识检索与事故预防的土木工程智能安全管理大语言模型框架[J].土木工程与管理学报,2026,43(1):108-117,10.基金项目
"十四五"国家重点研发计划(2022YFC3802703-04) （2022YFC3802703-04）
陕西省重点研发计划项目(2024GX-ZDCYL-02-04) （2024GX-ZDCYL-02-04）
