To address efficiency bottlenecks in traditional TRIZ function model construction, this study proposes an intelligent generation method leveraging large language models (LLMs). By developing a dual-layer functional ontology framework and designing multi-level prompt engineering strategies, we achieve automated conversion from natural language problem descriptions to structured TRIZ function models. Comparative experiments across mainstream LLM platforms demonstrate the system’s generation quality and response efficiency. Results indicate that the generated function models achieve high completeness and semantic accuracy across different strategy-platform combinations, while enabling dynamic parameter adjustment and multi-solution iteration. Practical engineering cases confirm the method’s effectiveness in reducing functional analysis barriers and providing real-time intelligent assistance, highlighting significant practical value for innovation design in complex systems.