| Keywords |
Digital twin; Building energy simulation; Multi-day calibration; Nelder-Mead |
| Abstract |
Building energy consumption continues to rise, increasing interest in digital twin?based calibration for improving operational efficiency. The Nelder?Mead algorithm has been widely applied to building energy models, yet prior studies have focused more on prediction accuracy and computational efficiency than on the physical reliability of calibrated parameters. This study compares multi-day indoor temperature calibration with the conventional single-day approach. A floor of an office building was examined under non-HVAC operating conditions using indoor temperatures measured from 11 sensors. After preprocessing, sensors with high correlation (Pearson > 0.8) were averaged to obtain a representative indoor temperature. Six parameters? U-value, SHGC, infiltration, ventilation, and two cross-mixing rates?were calibrated within extended but plausible boundaries. Single-day calibration yielded the lowest mean CV(RMSE) of 2.42 ± 1.16%, while same-season and cross-season multi-day calibration showed 4.30% ± 2.60% and 3.70% ± 0.60%, respectively. Although multi-day calibration showed slightly higher errors, all cases satisfied the ASHRAE Guideline 14 criterion. Single-day calibration also showed more frequent convergence to boundary values, whereas this tendency was reduced under multi-day calibration, especially in cross-season cases. These results suggest that multi-day calibration can alleviate boundary convergence while keeping calibration errors within the ASHRAE Guideline 14 threshold, although envelope-related parameters still converged to boundary values in some cases. |