Surrogate Based Global Optimization Using Adaptive Switching Infill Sampling Criterion

Abstract

A novel infill sampling criterion is proposed for a surrogate based global optimization algorithm. Due to the extensive amount of calculations required for the meta-heuristic global optimization methods, a surrogate model was employed. In the surrogate based global optimization, SBGO, an iterative process of constructing a model and sampling new points are repeated until a stopping criterion is met. An infill sampling criterion, ISC, controls which data point should be sampled, however, because the characteristics of a design problem are prone to influence the performance of the algorithm, an adaptive ISC should be developed. Thus, in this study, an algorithm that adaptively searches globally and locally considering the current existing samples is proposed. The novel ISC is integrated with a global search measure weighted minimum distance, WD, which considers not only the most ambiguous regions but also accounts for the response values for higher efficiency. The algorithm was tested on unconstrained mathematical functions including the Dixon-Szego test functions and the results were compared with other SBGO algorithms. Additionally, the algorithm was further expanded and implemented to constrained optimization problems using penalizing coefficients.