Exploring floor plan design to achieve indoor thermal comfort in public housing: An integrated heat graph and machine learning approach

Abstract

In densely populated regions, public housing constitutes a significant portion of residential buildings. Given its impact on public housing residents’ well-being, optimizing indoor thermal comfort (ITC) is important, especially in hot and humid climates. Traditional approaches to the optimization of floor plans rely heavily on designers’ experience and intuition. This research, in contrast, seeks empirical evidence for the impacts of floor plan design on ITC. It does so by employing graph theory to represent public housing floor plans, and then adopting the machine learning model XGBoost to interpret the complex interactions between graph variables and ITC indicators. We find that building layout considerations such as modularity, density, and connectivity have relatively more impact on ITC. Highly modular zoned designs, while improving overall comfort, often face challenges related to thermal stability. In contrast, low-density linear layouts in naturally ventilated environments can enhance comfort by improving airflow and heat dissipation. This research contributes to the development of efficient design strategies for public housing and provides an evidence-based research framework for future generative design.