Compensating errors in building energy simulation calibration: quantifying data-driven behavioral uncertainty

Building energy performance gap is partly attributed to simplified occupant behavior representations in simulations. Conventional calibration approaches satisfy statistical criteria but mask inconsistent physical assumptions through compensating errors. This study examines how stepwise integration of measured occupant behavior alters calibration outcomes, exposing compensating errors in conventionally calibrated models. Indoor temperature, humidity, presence, lighting, window-opening behavior, and weather were collected during a one-year monitoring campaign in an educational building. An iterative calibration framework was implemented using EDSL Tas (Environmental Design Solutions Ltd. TAS), progressing from an uncalibrated baseline to a statistically calibrated reference model and subsequently to behavior-integrated iterations. Although the conventionally calibrated model achieved strong statistical agreement with measured heating energy (NMBE = 1.39%), replacing standard schedules with measured presence increased predicted heating demand (3.96–4.19%). This revealed compensating errors associated with overestimated internal gains and constrained envelope parameters. Adaptive window-opening behavior shifted disagreement to 7.05% due to ventilation losses. Measured lighting use acted as a corrective internal gain, partially offsetting heating demand. Findings demonstrate that compliance with calibration benchmarks alone is insufficient to ensure physically accurate models. Monitored occupant behavior is essential for identifying compensating errors and establishing physically consistent baseline models suitable for evaluating energy conservation measures in existing buildings.