Neural network-enhanced BIM framework for designing thermally comfortable urban environments

초록

Outdoor thermal comfort (OTC) evaluation during the design phase is often compromised by the computational complexity of physical simulations. Consequently, current BIM-integrated workflows predominantly rely on simplified proxies, assuming surface temperature equals air temperature. This study hypothesizes that such simplifications introduce significant errors in long-wave radiation calculations, thereby distorting thermal comfort assessments. To address this, we present a computational framework that integrates a physics-based surrogate model (ESMUST), trained on 42 million EnergyPlus datasets, into the BIM environment. Unlike conventional methods, this framework predicts surface temperatures based on thermodynamic principles. A comparative analysis conducted on a university campus demonstrated that the proposed method captures surface temperature variations that proxy-based approaches inherently cannot represent. The results indicate that explicit surface temperature modeling alters the calculated COMFA thermal budget by 22.5-25.1 W/m2, a magnitude sufficient to shift thermal comfort classifications and influence design decision-making. Comparative validation with ENVI-met simulations confirmed consistent thermal comfort improvement patterns, with the proposed framework achieving computational times approximately two orders of magnitude faster. This research demonstrates the necessity of physics-based surface temperature modeling in BIM-integrated thermal comfort assessment and provides a computationally efficient methodology for early-stage design evaluation.

키워드