Research on natural language driven intelligent visualization generation based on deep integration of LLM and BIM

Building Information Modeling (BIM) drives AEC industry digitalization but is hindered by high operational complexity, while Large Language Models (LLMs) excel at natural language interaction yet suffer from domain knowledge gaps, hallucinations, and inability to generate executable BIM operations. To address these complementary bottlenecks, this study proposes a four-layer deep integration framework based on Knowledge-Enhanced LLM and Structured Workflow Mapping. The framework integrates a standardized BIM workflow knowledge base, an intelligent natural language-to-workflow mapping mechanism with fine-tuned LLaMA-2-7B and dual verification, an intelligent resource matching module, and an end-to-end system integration layer connecting mainstream BIM software. Rigorous experiments demonstrate the framework’s superiority, achieving 91.6% semantic accuracy, 86.2% workflow success rate, and 18.5- second average task completion time—outperforming general LLMs and manual operations. This study clarifies LLM-BIM complementarity, bridges the natural language-BIM semantic gap, and promotes digital design democratization in the AEC industry, providing a feasible solution for intelligent BIM applications.