Comparative evaluation of grey-box virtual sensor under different indoor thermo-hydrodynamics in room heating

• Grey-box virtual sensor model was evaluated under two indoor flow conditions. • Selection of online data location (X) is critical to model accuracy. • X should avoid forced convection from the supply jet during HVAC. • Model accuracy tends to benefit from proximity between X and the target location. • Corner locations are not suitable as X owing to weak convection characteristics. Indoor temperature is a critical parameter in building control systems, affecting both energy efficiency and thermal comfort, and grey-box models have been proposed to predict it. A promising grey-box model has been previously developed to infer indoor interior temperatures by referencing online temperature at a wall location. This study examined how indoor thermo-hydrodynamics influence this model’s performance and explored strategies to improve its predictive accuracy. Water-based experiments were conducted using two nozzle configurations to simulate representative flow fields in room heating processes. Modeling and cross-validations were conducted using the collected data. Flow dynamics were analyzed using particle image velocimetry (PIV) to elucidate the underlying mechanisms. Comparative analyses revealed that the reference location is critical to model accuracy owing to convective influences. Therefore, alternative reference locations were tested, revealing substantial variations in accuracy. Particularly, the prediction accuracy was substantially improved by 61%, with mean absolute errors reduced from 1.35 K to 0.53 K in the Y-shaped nozzle case. Thermal correlation analyses further revealed that flow dynamics can substantially alter the thermal correlations among the target, reference, and nozzle locations in a room. Based on these findings, three practical guidelines for selecting a reference location were established: (1) avoid supply jet convection, (2) consider target proximity, and (3) avoid corner locations. These findings can facilitate the deployment of the virtual sensor model with physical insights, contributing to intelligent indoor temperature control and energy-efficient HVAC operations.