Demonstrating the feasibility of LLMs to develop complex building energy models

Decarbonizing the existing building stock is a critical challenge. Current Building Energy Modeling (BEM) processes are time consuming and costly, limiting their scalability. Large language models (LLMs) can expedite BEM, yet challenges remain in determining effective prompts that incorporate necessary spatial information when modeling complex buildings.

Our research conducted a comparative analysis between traditional and LLM-driven BEM, demonstrating that LLM-driven BEM can achieve up to an overall 93.6% match to an industry-standard calibrated model. Our findings highlight the potential of LLM-driven BEM to reduce the need for specialized training and allow wider application to the entire building stock, facilitating decarbonization.