This study aims to comprehensively assess the transferability of optimized values of independent design variables from an early-stage single-storey model to a detailed whole building and to identify the most influential design variables to reduce thermal discomfort, cooling energy and improve daylight for a residential building in the warm and humid region of Chennai.
This study consists of six phases. In the initial phase, the simulation results were validated using field measurements. Four design cases were developed for the early to detailed design stages, and multi-objective optimization (MOO) was performed for each design case. Multiple regression analysis was performed to identify the most significant (p < 0.05) independent design variables for the performance objectives, and the magnitude of their influence was measured using the standardized coefficient beta (ß). A deep learning sequential model was built for each design case, and the overall influence of the design variables was analyzed using permutation feature importance.
The results revealed that the cooling setpoint, energy efficiency of the air conditioner (ISEER) and window-to-wall ratios (N-S-E-W) exhibited greater transferability and variations in importance from the early to detailed design stages. Most of the independent design variables considered in this study showed both statistical significance and non-significance across the design cases.
This study provides guidance for architects and engineers by emphasizing prioritizing the most influential independent design variables for achieving high-performance residential buildings.
This study combines statistical measures and permutation feature importance to highlight the sensitivity of design variables from the early to detailed design stages.
