基于大语言模型的BIM正向设计问答系统研究

Research on LLM-based Question-Answering System for BIM Forward Design

丁志坤 ¹李金泽 ²刘明辉³

扫码查看

作者信息

1. 深圳大学滨海城市韧性基础设施教育部重点实验室,广东深圳 518060;深圳大学深圳市地铁地下车站绿色高效智能建造重点实验室,广东深圳 518060;人工智能与数字经济广东省实验室(深圳),广东深圳 518107
2. 深圳大学中澳BIM与智慧建造研究中心,广东深圳 518060
3. 深圳大学土木与交通工程学院,广东深圳 518060
折叠

摘要

在正向设计中,设计人员能够通过BIM技术进行设计和管理建筑项目,以支持决策.然而,BIM正向设计过程中的信息处理和知识共享仍面临诸多挑战,尤其是在快速获取和应用复杂建筑规范及设计知识方面.为解决上述问题,本研究基于大语言模型(LLM)设计了一套BIM正向设计问答系统,旨在探究LLM在BIM正向设计场景中的实际应用.通过采用基于QLoRA的LLM微调方式提高模型的专业性,并构建本地知识库形成问答系统,最后设计评价指标对系统进行主客观性能评估、专业问答对比分析.实验结果表明,该系统在专业领域性能表现良好,并能根据特定设计场景生成建议和解决方案,验证了本系统的可行性和优越性.

Abstract

In forward design,designers can design and manage construction projects by BIM technol-ogy to support decision-making.However,the process of information processing and knowledge sha-ring in BIM forward design still faces many challenges,especially in terms of quickly accessing and applying complex construction regulations and design knowledge.To address these issues,this study designed a BIM forward design Q&A system based on large language model(LLM),aiming to explore the practical application of LLM in BIM forward design scenarios.By adopting a QLoRA-based LLM fine-tuning approach,the model professionalism is enhanced and a local knowledge base is built to develop the Q&A system.Finally,the system is evaluated using both subjective and objective perform-ance metrics,and professional Q&A comparative analysis.Experimental results show that the system performs well in the professional domain by generating suggestions and solutions for specific design scenarios,which validate its feasibility and superiority.

关键词

正向设计/BIM/大语言模型/专业问答

Key words

forward design/BIM/large language model/professional Q&A

引用本文复制引用

基金项目

深圳市高等院校稳定支持计划重点项目(20220810160221001)

出版年

2024

土木工程与管理学报

华中科技大学

土木工程与管理学报

CSTPCDCHSSCD

影响因子：0.837

ISSN：2095-0985

参考文献量19

段落导航