当前建筑信息模型（Building Information Modeling，BIM）技术在建筑领域得到了广泛应用，推动建筑行业向智能化方向发展，但仍存在建模自动化程度低、与设计分析阶段未能完全融合等诸多局限性。大语言模型和以深度学习为代表的人工智能技术凭借其强大的自然语言理解能力以及高效性、通用性和可迁移性，在土木结构设计和计算领域拥有着巨大的应用潜力。本论文基于大语言模型（Large Language Models，LLMs）、二次开发技术和图神经网络模型，主要研究了BIM软件内大语言模型-建筑信息模型-智能计算技术的一站式工作流建立方法，论文的主要研究成果如下：（1）实现基于LLM的交互式建筑结构设计。通过构建多代理系统（Multi-Agent System，MAS）框架实现了使用者自然语言到钢筋混凝土（Reinforced Concrete，RC）框架结构BIM模型的自动化建模方法。多代理框架为代理提供了程序集接口和基于模型检查器(Model Checker，MC)的循环检查流程以保证模型生成质量。评估结果表明虽然该框架仍存在不足与局限性，但已经可以稳定有效生成高质量、结构合理的简单RC框架结构模型。（2）实现BIM模型自动化配筋算法与计算模型转换。基于Revit软件二次开发实现RC框架结构的参数化和自动化配筋方法。基于C#和Python二次开发实现Revit模型向有限元软件SAP2000计算模型的转化方法，通过实验证明转化模型的准确性。（3）实现框架结构的智能计算方法与一站式计算平台构建。提出了适用于框架结构体系描述的同构图数据结构并构建了混凝土框架结构体系数据集。开发理论驱动的图神经网络模型，进行简单框架结构体系弹性分析，与有限元结果对比验证了部分内力计算结果的准确性。在Revit系统中开发构建一站式计算平台，进行计算结果的实时反馈和模型设计变更，实现结构自动生成到智能计算的一站式解决方案。

Building Information Modeling (BIM) has been widely applied in the construction industry, driving the sector towards intelligent development. However, it still faces several limitations, such as low automation in modeling and incomplete integration between the design and computation phases. Large language models (LLMs) and artificial intelligence technologies, particularly those represented by deep learning, have enormous potential in civil structural design and computation due to their natural language understanding capabilities, efficiency, versatility, and transferability. This paper, based on large language models, BIM secondary development technology, and graph neural network models, primarily investigates a one-stop workflow for integrating LLM, Building Information Modeling (BIM), and intelligent computation technologies within Revit. The main research findings of this paper are as follows:(1) Implementation of interactive building structural design based on LLM. An automated modeling method for reinforced concrete frame structures in BIM using natural language input is achieved through the development of a multi-agent system. The framework provides agents with an assembly interface and a Model Checker (MC)-based iterative check process to ensure the quality of the generated models. The evaluation results show that, although the framework still has some limitations, it can effectively and stably generate high-quality, structurally reasonable simple RC frame models.(2) Implementation of automated reinforcement design algorithms and transformation of computational models in BIM. A parametric and automated reinforcement design method for RC frame structures is realized through Revit secondary development. A method for converting Revit models into finite element models for SAP2000 is developed using C# and Python, with experimental results demonstrating the accuracy of the transformed models.(3) Implementation of intelligent computation methods for frame structures and development of a one-stop computational platform. A heterogeneous graph data structure suitable for describing frame structures is proposed, and a concrete frame structure dataset is created. A theory-driven graph neural network model is developed for the elastic analysis of simple frame structures, and its results are compared with finite element analysis to verify the accuracy of internal force calculations. A one-stop computational platform is developed within the Revit system to provide real-time feedback on computational results and model design changes, thus realizing a one-stop solution for automated generation and intelligent computation of structures.