From design to operation: Multi-agent AI for virtual in-situ modeling of digital twins in BIM

A virtual building model (VBM) is a mathematical representation that describes the behavior of a physical building. Accurate VBMs provide insightful information about the physical building within the digital twin (DT) framework. However, there is limited research on the autonomous construction of VBMs. To address this gap, this paper proposes a multi-agent artificial intelligence (AI) system that autonomously constructs VBM. The proposed system autonomously develops, calibrates, and manages virtual models that constitute the VBM by leveraging data and information generated throughout the building lifecycle within a BIM environment. The proposed method is validated on a real operating heating, ventilation, and air-conditioning (HVAC) system, autonomously developing a chilled water flow rate model (MAPE 2.27 %) and an evaporator inlet temperature model (RMSE 0.33°C). These results suggest the feasibility of autonomously constructing VBMs and contribute to shifting the DT paradigm from physical construction automation to the autonomous construction of VBM.