Interpretable reduced-order optimization research for improving indoor environmental quality in buildings based on dimensional analysis and surrogate-based optimization

Abstract

Implementing large-scale optimization designs and identifying optimal design variables in fields such as architectural engineering is crucial for improving indoor environmental quality. However, the high-dimensional nonlinear characteristics complicate the understanding of the optimization process and increase the difficulty of finding optimal solutions. This study proposes an interpretable reduced-order optimization (ROO) method based on dimensional analysis, which integrates numerous variables into a single reduced-order coefficient through a power form. The goal is to develop a reduced-order and approximately one-dimensional nonlinear relationship during the optimization process and to reveal the mechanisms of single-objective and multi-objective optimization, thereby enhancing the interpretability of the optimization process. The reliability of the ROO is validated through single-objective optimization of building thermal bridges and multi-objective optimization of squirrel cage fans. The study shows that the power in the reduced-order coefficient can be used to determine the sensitivity and impact of design variables on optimization targets. The nonlinear relationship between the composite structural parameters of the building thermal bridge and the total heat flux can be reduced to an approximately one-dimensional linear relationship, significantly reducing the total heat flux of the thermal bridge by 10.3 %. The ROO accurately describes the aerodynamic performance of the fan and its high-dimensional nonlinear relationship with eight design variables, eliminating optimization conflicts among multiple operating conditions and significantly enhancing the aerodynamic performance. The ROO method proposed in this study offers a new approach to making the optimization process more interpretable, playing a crucial role in optimization design and revealing optimization mechanisms in engineering applications.