Application of Surrogate Models for Building Envelope Design Exploration and Optimization | www.esteco.com

Application of Surrogate Models for Building Envelope Design Exploration and Optimization

Author(s): 
Ding Yang, Yimin Sun, Rusne Sileryte, Antonio D’Aquilio and Michela Turrin (Delft University of Technology)

CHALLENGE - Building performance simulations are usually time-consuming, accounting for the major portion of time spent in Computational Design Optimization (CDO). Therefore, optimization may become less efficient or even infeasible within a limited time frame of real-world projects, due to the computationally expensive simulations. This research aims to investigate the potentials of surrogate models (i.e. Response Surface Methodology - RSM) to be used in the building envelope design exploration and optimization.

SOLUTION -  This work investigates how, and to what extent problem scales may affect the application of RSM, and how different ways of using RSM may affect the quality of Pareto Front approximations. A series of multi-objective optimization tests are carried out. For the application of RSM, a simple building envelope design optimization was formulated in Grasshopper. The two buildings are assumed to be one-story sports halls with a fixed rectangular plan and a changeable spherical roof, located in South China. Energy Use Intensity and Illuminance Uniformity (IU) are selected as performance criteria, while Spatial Daylight Autonomy (sDA)and Average Illuminance are chosen as performance constraints. Therefore, annual hourly daylight and energy simulations are performed by Daysim and Energyplus sequentially. The platform couples Grasshopper with modeFRONTIER.  In order to develop the surrogate model, the following steps are followed: 1) Sampling of Design Points by DoE; 2) Data Collection by Running Simulations; 3) Surrogate Model Generation by RSM.

 

 

 

 

 

BENEFITS - In order to investigate possible effects of problem scales, the accuracy of surrogate models in the two proposed cases are compared. In general, the accuracy of surrogate models in Case 1 appears to be better than that in Case 2. Current results suggest that the IU and sDA estimation in Case 1 is much better than that in Case 2. 

LOG IN TO DOWNLOAD