Online selection of surrogate models for constrained black-box optimization
Samineh Bagheri, Wolfgang Konen, Thomas Bäck · 2016
Real-world optimization problems are often subject to many constraints which are expensive to be evaluated in terms of cost and time. Surrogate-assisted optimization techniques aim to reduce evaluation costs by substituting the objective and the constraint functions with cheap surrogate models e. g. radial basis functions. Selecting the correct basis function and its associated parameters is a challenging task and depends on many factors like initial design, population size and type of function to be modeled, which is usually unknown in black-box problems. We propose an online model selection strategy that sorts different models according to their approximation error and selects in each iteration the best model for each function. The model selection strategy is embedded in the SACOBRA optimizer [1]. We show that the proposed online model selection strategy boosts the overall performance of the algorithm when applied to 24 well-known constrained optimization problems.