Enhancing simulation accuracy in building energy modeling through data-driven approaches

This thesis investigated the contribution of occupant behavior towards residential building energy consumption by comparing deterministic and probabilistic schedule models. 170 in-depth survey responses were obtained across Türkiye in an effort to record daily residential activities every 15 minutes. These were augmented into 1000 high-resolution daily occupant schedules with the incorporation of variation in behavior into energy simulations. Two residential building models, a high-rise and a low-rise configuration were simulated using Energy Plus with fixed (deterministic) and variable (probabilistic) schedule methods. Importantly, the occupant schedules used in both models were identical; the only difference between the two scenarios was the building form, allowing analysis of geometry-driven energy variations. The methodology used consisted of realistic probabilistic schedule creation using MATLAB and Python, automated interfacing with EnergyPlus as CSV inputs, and simulation of 50 randomized runs per scenario. The deterministic models built on standard daily routines from the literature and duplicated over all days of the year. The outputs of the simulations were evaluated in five categories of energy consumption: lighting, HVAC, other electrical uses, total electricity, and total utility consumption. The outcomes revealed that probabilistic values tend to occur around the average of probabilistic distributions but could not capture extreme behaviors that play a significant role in system sizing and peak load. v Probabilistic models had wider variability in plug loads and electricity consumption but less varied HVAC loads that still remained influenced by changing patterns of occupant presence. The results highlighted the necessity for real occupant behavior to be included within building performance simulation for better energy demand representation. Total average energy usage for probabilistic simulation ranged between 63.9–79.5 kWh/m² for the two scenarios, compared to 74.2 and 71.4 kWh/m² under deterministic values. Variability was seen to be restricted for loads under HVAC, but varied considerably for other plug loads and lighting based on different behavior patterns. These observations reinforce the fact that internal variation is hidden under deterministic modeling, and that probabilistic simulation gives better insight into actual occupant impact on energy usage. The research brought a culturally informed, fact-based modeling approach applicable in Turkish residential buildings and confirmed that probabilistic simulation methods offer a stronger and more realistic platform for analyzing the energy supply and demand, evaluation of policies, and optimization of sustainable designs.