Integrating Context-Dependent Occupant Behavior into Urban Building Energy Modeling: A Data-Driven Framework for High-Density Housing

Accurate prediction of building energy consumption is critical for sustainable urbanization, while dynamic occupant interactions with the building envelope remain a major source of uncertainty. Prevailing cross-scale building energy models often rely on simplified or uniform behavioral assumptions that cannot adequately represent the spatial and temporal heterogeneity of occupant actions such as window opening and curtain use. This study proposes an AI-augmented framework that integrates field-observed occupant behavior, machine-learning prediction, and physics-based building energy simulation. Time-series observations of window and curtain states were collected from 1609 rooms across 12 residential buildings in a high-density neighborhood in Hong Kong. Environmental and contextual associations were first examined at the aggregated behavioral level, after which Random Forest models were used to generate dynamic behavior schedules. These schedules were subsequently integrated into EnergyPlus to evaluate how different levels of occupant behavior representation affect simulated cooling energy demand and computational cost. Compared with the Detailed Scenario, the Template Scenario produced 20.5% higher simulated cooling energy, while the Average Scenario showed an 8.7% relative difference with 20.6% less simulation time. An additional controlled analysis showed that removing surrounding buildings increased simulated cooling energy by 1.7–9.3%, demonstrating the direct physical influence of inter-building shading. Overall, the proposed framework provides a practical pathway for incorporating context-dependent occupant behavior into urban building energy modeling and for evaluating the trade-offs between behavioral modeling detail and computational efficiency.