A large language model-based framework with retrieval-augmented generation for automated building energy modeling

Building energy models (BEMs) are crucial for design and analysis but often require considerable time and expertise due to reliance on specialized software and manual configuration. To address these challenges, this study proposes a framework based on large language model (LLM) with retrieval-augmented generation for automating building energy modeling. The proposed framework leverages the advanced natural language processing capabilities of LLM to parse user inputs expressed in natural language. Historical BEMs that match these requirements are retrieved from a feature database, providing reference data to the LLM to enhance the accuracy and adaptability of the generated results. This approach enables the rapid construction of BEMs tailored to user specifications. Experimental results demonstrate that the modeling time is reduced from several hours, which is typical of traditional manual methods, to just a few minutes, resulting in a significant improvement in efficiency. Furthermore, the accuracy of the load simulations is highly consistent with the results from manual modeling, confirming the reliability of this method in real-world applications. This framework offers an efficient and intelligent solution for building energy analysis and simulation, highlighting the substantial potential of large language model in advancing building simulation and performance modeling.