Novel composition test functions for numerical global optimization

Abstract

In the evolutionary optimization field, there exist some algorithms taking advantage of the known property of the benchmark functions, such as local optima lying along the coordinate axes, global optimum having the same values for many variables and so on. Multiagent genetic algorithm (MAGA) (1) is an example for this class of algorithms. In this paper, we identify shortcomings associated with the existing test functions. Novel hybrid benchmark functions whose complexity and properties can be controlled easily, are introduced and several evolutionary algorithms are evaluated with the novel test functions.