A Study on GPT-assisted Virtual Model Construction Using BIM

A study on GPT-assisted virtual model construction using BIM

This study proposes a method that uses GPT to autonomously extract and parameterize BIM data for automating virtual modeling. Physical equations are generated from BIM data in the design phase to create an initial virtual model, which serves as a foundation for the expansion of the Digital Twin (DT). Additionally, a system architecture is introduced to automate the modeling and performance calibration of the virtual model, supporting the continuous expansion and synchronization of the DT. This approach was applied to the chilled water flow in an HVAC system, and the calibrated model, using operational data, achieved a high accuracy with a MAPE of 2.5%, demonstrating the potential to improve smart building energy systems.