We have employed a recent implementation of genetic algorithms to study arange of standard benchmark functions for global optimization. It turns outthat some of them are not very useful as challenging test functions, since theyneither allow for a discrimination between different variants of geneticoperators nor exhibit a dimensionality scaling resembling that of real-worldproblems, for example that of global structure optimization of atomic andmolecular clusters. The latter properties seem to be simulated better by twoother types of benchmark functions. One type is designed to be deceptive,exemplified here by Lunacek's function. The other type offers additionaladvantages of markedly increased complexity and of broad tunability in searchspace characteristics. For the latter type, we use an implementation based onrandomly distributed Gaussians. We advocate the use of the latter types of testfunctions for algorithm development and benchmarking.
展开▼
