Impacts of problem scale and sampling strategy on surrogate model accuracy

Impacts of problem scale and sampling strategy on surrogate model accuracy: An application of surrogate-based optimization in building design

Yang, D. (TU Delft Design Informatics; South China University of Technology)
Sun, Y (South China University of Technology)
Di Stefano, D. (ESTECO SpA)
Turrin, M. (TU Delft Design Informatics; South China University of Technology)
Sariyildiz, I.S. (TU Delft Design Informatics; Yasar University)

2016-11-21

Surrogate-based Optimization is a useful approach when the objective function is computationally expensive to evaluate, compared to Simulation-based Optimization. In the surrogate-based method, analytically tractable “surrogate models” (also known as “Response Surface Models — RSMs” or “metamodels”), are constructed and validated for each optimization objective and constraint at relatively low computational cost. They are useful for replacing the time-consuming simulations during the optimization; quickly locating the area where the optimum is expected to be for further search; and gaining insight into the global behavior of the system. Nevertheless, there are still concerns about the surrogate model accuracy and the number of simulations necessary to get a reasonably accurate surrogate model. This paper aims to unveil: 1) the possible impacts of problem scale and sampling strategy on the surrogate model accuracy; and 2) the potential of Surrogatebased Optimization in finding high quality solutions for building envelope design optimization problems. For this purpose, a series of multi-objective optimization test cases that mainly consider daylight and energy performance were conducted within the same time frame. Then, the results were compared, in pair, based on which discussions were made. Finally, the corresponding conclusions were obtained after the comparative study.

multi-objective optimization
surrogate-based optimization
problem scale
sampling strategy
response surface model
design of experiments

http://resolver.tudelft.nl/uuid:d7c06083-3b1c-4d3f-a209-8984569c4a46

https://doi.org/10.1109/CEC.2016.7744323

IEEE

978-1-5090-0622-9

2016 IEEE Congress on Evolutionary Computation (CEC)

2016 IEEE Congress on Evolutionary Computation, CEC 2016, 2016-07-24 → 2016-07-29, Vancouver, Canada

Accepted Author Manuscript

Institutional Repository

conference paper

© 2016 D. Yang, Y Sun, D. Di Stefano, M. Turrin, I.S. Sariyildiz