A comparison of conventional and meta-model based global optimization methods

Motivated by the growing number of applications in engineering, physics, science and other fields, interest in the development of global optimization algorithms is increasing. In this paper, two categories of global optimization methods are considered, namely conventional and meta-model based algorithms. Conventional algorithms require values of the objective function to obtain a solution, while meta-model based algorithms can be used with incomplete information or when there is a limit on the available time or cost. Complex functions pose a challenge to gradient-free algorithms as they may need a significant number of function evaluations, thus meta-model based techniques may be preferred. In the paper, these algorithms are compared using a set of benchmark problems which include convex and non-convex problems, as well as smooth and non-smooth problems.