Synergizing Design of Building Energy Performance Using Parametric Analysis, Dynamic Visualization, and Neural Network Modeling

Abstract

Designing buildings is a multicriteria decision-making process that usually involves a large number of design parameters and several objective functions. The associated combinatorial parametric sensitivity analysis requires numerous simulation runs, which might not be practical, feasible, or both. The parameters related to the design of buildings include geometry and envelope characteristics, uncertainty in internal loads, different HVAC system characteristics, and utility rate structures. A new methodology is proposed that involves three stages: (1) an initial one-factor-at-a-time (OAT) statistical method (Morris method), which is very efficient when identifying the relative importance and interactivity of parameters; (2) the use of parallel coordinates and other graphical plots to help visualize ascertained allowable latitude of parameters dynamically and interactively; and (3) the use of a machine learning algorithm [specifically, artificial neural networks (ANN)] to include the improved granular domain of parameters. This results in more flexibility when exploring the design space and reducing the number of computationally intensive simulation runs without compromising the mathematical resolution and accuracy. The method would empower designers to explicitly analyze the impacts of all major influencing input parameters while providing flexibility to posit different constraints on selected parameters and visualize their interaction with other parameters. In addition, it has advantages over traditional optimization approaches since decisions can be made by assessing and controlling one or more objective functions (response variables or evaluation criteria) and input parameters simultaneously under preset bounds. This is especially useful when there are multiple objective functions that are conflicting. The various stages of the proposed methodology are demonstrated through a hypothetical building design study.