1 College of Civil Engineering and Transportation, Northeast Forestry University, Harbin 150040, China
2 National Center of Technology Innovation for Green and Low-Carbon Building, School of Energy and Civil Engineering, Harbin University of Commerce, Harbin 150028, China
Show less
文章历史+
Received
Published
Revised
2025-07-28
2026-05-28
2025-12-16
Online Date
2026-08-07
PDF (16297K)
摘要
建筑施工安全标准因其非结构化特性难以满足现代建筑施工安全管理需求。大语言模型(large language model,LLM)具备强大的自然语言处理能力,但在专业领域的“幻觉”问题制约其在建筑施工安全场景中的应用。将图检索增强生成(GraphRAG)技术与LLM结合,构建面向建筑施工安全的增强大语言模型智能信息检索与问答系统。首先构建了建筑施工安全标准知识图谱,实现了标准文本的结构化表达,并基于深度学习方法对提问语句进行命名实体识别与匹配,实现对用户提问的实体级语义理解与知识定位。通过提示工程与LLM交互,有效提升模型回答准确性。最后系统试运行验证了其在多种建筑施工场景中的有效性。研究结果表明,该系统有助于提高建筑施工安全管理信息化水平,推动建筑施工安全标准知识的智能化应用。
Abstract
Owing to their unstructured format, construction safety standards struggle to meet the demands of modern construction safety management. Although large language models (LLMs) possess strong natural language processing capabilities, their use in specialized fields such as construction safety is hindered by the problem of “hallucination”. Combining graph retrieval-augmented generation (GraphRAG) with LLMs, an intelligent information retrieval and question answering system based on the augmented large language model for construction safety was constructed. A knowledge graph of construction safety standards was first built to achieve a structured representation of standard texts. Subsequently, a deep learning-based approach was leveraged to conduct named entity recognition and matching of query statements, facilitating entity-level semantic comprehension and knowledge localization. By integrating prompt engineering with LLM interaction, the accuracy of model responses was effectively improved. Trial runs of the system confirm its effectiveness across diverse construction scenarios. The findings suggest that this system contributes to elevating the informatization of construction safety management and advances the intelligent utilization of construction safety standard knowledge.
SU Yi-kun, LIU Jun-hao, ZHENG Zhi-zhe, WANG Xing, YANG Xiao-zhuang.
Enhanced Large Language Model-based Intelligent Information Retrieval and Question-answering System for Construction Safety[J].
Science Technology and Engineering, 2026, 26(15): 6659-6670 DOI:10.12404/j.issn.1671-1815.2505554
语言模型(language model,LM)被定义为一种用于建模自然语言的概率模型,其根据前一个标记的概率分布来预测序列中的下一个标记,其主干为深度神经网络模型[9]。早期模型如循环神经网络和长短期记忆网络在捕捉长期依赖关系和实现并行计算等方面面临挑战[12]。为此,Transformer模型通过引入注意力机制用以解决这些问题,以更加有效的方式对输入文本进行上下文表示[13]。这一创新为双向变换器模型(bidirectional encoder representation from transformers,BERT)和生成预训练转换器等先进模型提供了重要基础,允许它们利用大规模预训练来开发模型,使得机器可以更好地理解自然语言。大语言模型就此诞生,即通过大量文本数据训练的深度学习模型。
LLM具有先进的语言理解能力和处理各种自然语言处理任务的能力,如智能问答、文本摘要等,成为处理复杂知识的强大工具[14-16]。LLM可以在海量的文本数据集上进行训练,使得它们可以回答广泛的查询任务,无需费力收集和注释特定领域数据[17]。在建筑领域,研究人员通过大语言模型技术为许多问题的解决提供了方案。例如,Zheng等[18]通过将文本转化为结构化查询语言(structured query language,SQL),利用LLM从建筑信息模型(building information modeling,BIM)中查询信息,为项目管理提供建议。Lin[19]利用LLM技术为建筑施工项目的现场管理人员开发了一个聊天机器人,帮助他们更好地对施工现场进行管理。Wu等[20]构建了一个用于施工管理的增强信息驱动的信息检索与问答系统,以从项目文档中检索准确信息。在建筑施工安全领域,张朋杨等[21]和Smetana等[22]等利用LLM在安全法规中检索信息,总结事故报告,分析建筑工伤的原因。董磊等[23]提出一种基于多模态大语言模型的多模态知识图谱构建方法,为建筑施工现场的安全管理决策和智能推理提供了重要支持。
以建筑施工安全标准知识为研究对象,相关知识来源于住房和城乡建设部和国务院发布及主管的相关标准、规范和条例。通过全国标准信息公共服务平台、中华人民共和国住房和城乡建设部网站和工程建设标准化信息网进行收集,收集结果如表1所示。虽然相关标准、规范和条例是公开披露和可获取的,但文件通常是以PDF格式文件提供,因此使用光学字符识别(optical character recognition,OCR)技术进行内容提取,并使用TXT文本进行存储。
从提问语句中提取出实体及其概念后,通过语义相似性搜索映射到建筑施工安全标准知识图谱中相关实体和概念。使用word2vec方法进行语义相似性检验。数据集采用Chinese Word Vectors中文词向量数据集,该数据集由中国人民大学和北京师范大学学者开发,数据集提供超过100种中文词向量,具有多种表示方式和上下文特征[36]。计算提问语句中提取的实体及其概念与建筑施工安全标准知识图谱中的实体和概念的语义相似性分数,用以获得最佳匹配项。word2vec通过计算两个词的余弦相似度来衡量相似性,其计算公式为
DingMingzhi, BaiYonglan. Measurement and evaluation of work safety level in China’s construction industry[J]. Journal of Safety and Environment, 2024, 24(5): 1740-1748.
ChangChunguang, LingXiaoxue. Optimization of safety investment in prefabricated building construction based on sparrow search algorithm optimized support vector regression[J]. Science Technology and Engineering, 2023, 23(35): 15328-15334.
[5]
ChanC F, WongP K Y, GuoX, et al. Context-aware vision-language model agent enriched with domain-specific ontology for construction site safety monitoring[J]. Automation in Construction, 2025, 177: 106305.
[6]
HuZ, LiS, HeC, et al. Literature review on construction safety resilience: a bibliometric analysis for mapping the state-of-the-art[J]. Safety and Health at Work, 2025, 16(3): 259-267.
LiuJunhao, SuYikun, ZhangZhibo, et al. Research on standards knowledge support for construction safety[J]. Journal of Safety Science and Technology, 2024, 20(7): 156-162.
YangYuexiang, TuXinyu, LiuWenling. Research on the construction and application of standard documents knowledge graph[J]. Digital Library Forum, 2022(6): 22-30.
[13]
ChenY, LuG, WangK, et al. Knowledge graph for safety management standards of water conservancy construction engineering[J]. Automation in Construction, 2024, 168: 105873.
[14]
ZhengC, WongS, SuX, et al. Automating construction contract review using knowledge graph-enhanced large language models[J]. Automation in Construction, 2025, 175: 106179.
[15]
AmugnogoL M, MascheroniP, BrooksS, et al. Retrieval augmented generation for large language models in healthcare: a systematic review[J]. PLoS Digital Health, 2025, 4(6): e0000877.
[16]
JeonK, LeeG. Hybrid large language model approach for prompt and sensitive defect management: a comparative analysis of hybrid, non-hybrid, and GraphRAG approaches[J]. Advanced Engineering Informatics, 2025, 64: 103076.
[17]
ZhuZ, HuangT, WangK, et al. Graph-based approaches and functionalities in retrieval-augmented generation: a comprehensive survey[J/OL]. (2025-04-08)[2025-12-16]. https://dl.acm.org/doi/10.1145/3795880.
[18]
SuY, ShiL, ZhouK, et al. Knowledge-informed deep networks for robust fault diagnosis of rolling bearings[J]. Reliability Engineering & System Safety, 2024, 244: 109863.
[19]
TangL, SunZ, IdnayB, et al. Evaluating large language models on medical evidence summarization[J]. npj Digital Medicine, 2023, 6(1): 158.
[20]
ShauZ, YuZ, WangM, et al. Prompting large language models with answer heuristics for knowledge-based visual question answering[C]// 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2023: 14974-14983.
QiLizhong, RongJingguo, ZhangSu, et al. Expert system for BIM 3D design in power transmission and transformation engineering supported by large language model[J]. Electric Power Construction, 2025, 46(11): 47-57.
[23]
OuyangL, WuJ, JiangX, et al. Training language models to follow instructions with human feedback[J]. Advances in Neural Information Processing Systems, 2022, 35: 27730-27744.
[24]
ZhengJ, FischerM. Dynamic prompt-based virtual assistant framework for BIM information search[J]. Automation in Construction, 2023, 155: 105067.
[25]
LinW Y. Prototyping a chatbot for site managers using building information modeling (BIM) and natural language understanding (NLU) techniques[J]. Sensors, 2023, 23(6): 2942.
[26]
WuC, DingW, JinQ, et al. Retrieval augmented generation-driven information retrieval and question answering in construction management[J]. Advanced Engineering Informatics, 2025, 65: 103158.
ZhangPengyang, ShengLong, WangWei, et al. Construction of a mine accident knowledge graph based on large language models[J]. Journal of Mine Automation, 2025, 51(2): 76-83, 105.
[29]
SmetanaM, SallesD E, SukharevI, et al. Highway construction safety analysis using large language models[J]. Applied Sciences, 2024, 14(4): 1352.
DongLei, WuFuju, ShiJianyong, et al. Construction and application of multimodal knowledge graph in construction safety field based on large language model[J]. Computer Engineering and Applications, 2025, 61(9): 325-333.
[32]
LiuP, QianL, ZhaoX, et al. Joint knowledge graph and large language model for fault diagnosis and its application in aviation assembly[J]. IEEE Transactions on Industrial Informatics, 2024, 20(6): 8160-8169.
[33]
ChangT A, BergenB K. Language model behavior: a comprehensive survey[J]. Computational Linguistics, 2024, 50(1): 293-350.
[34]
ArslanM, MahdjoubiL, MunawarS. Driving sustainable energy transitions with a multi-source RAG-LLM system[J]. Energy and Buildings, 2024, 324: 114827.
XuChun, SuMingyu, SunBin. Tourism knowledge graph construction based on ChatGLM and prompt-tuning[J]. Science Technology and Engineering, 2024, 24(31): 13484-13492.
[37]
PengC, XiaF, NaseriparsaM, et al. Knowledge graphs: opportunities and challenges[J]. Artificial Intelligence Review, 2023, 56: 13071-13102.
[38]
DangL D, PhanU T P, NguyenN T H. GENA: a knowledge graph for nutrition and mental health[J]. Journal of Biomedical Informatics, 2023, 145: 104460.
ZhangZhanyong, WeiHaotian, DongShaohua. Risk early warning analysis of knowledge graph of accident evolution of complex gas pipeline network[J]. Science Technology and Engineering, 2025, 25(31): 13651-13659.
QinLi, HaoZhigang, LiGuoliang. Construction and correlation analysis of national food safety standard graph[J]. Journal of Computer Applications, 2021, 41(4): 1005-1011.
[43]
WuZ, MaG. NLP-based approach for automated safety requirements information retrieval from project documents[J]. Expert Systems with Applications, 2024, 239: 122401.
ShiHuiyang, WeiJingxuan, CaiXingye, et al. Research on threat intelligence extraction and knowledge graph construction technology[J]. Journal of Xidian University, 2023, 50(4): 65-75.
[46]
WuW, WenC, YuanQ, et al. Construction and application of knowledge graph for construction accidents based on deep learning[J]. Engineering, Construction and Architectural Management, 2023, 32(2): 1097-1121.
[47]
WangH, XuS, CuiD, et al. Information integration of regulation texts and tables for automated construction safety knowledge mapping[J]. Journal of Construction Engineering and Management, 2024, 150(5): 1-14.
[48]
LiS, ZhaoZ, HuR, et al. Analogical reasoning on chinese morphological and semantic relations[C]// Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics. Melbourne: Association for Computational Linguistics, 2018: 138-143.
QianFengdan, JiaTingting, ZhangDei, et al. Preliminary exploration of the application of the DeepSeek-V3-0324 large-scale model in medication education in pharmaceutical outpatient clinics[J]. China Pharmacy, 2025, 36(17): 2192-2196.