An Evolutionary Algorithm for Optimizing Functions with UV Structures

The function optimization is one of the most important optimization problems. In approaches to function optimization by evolutionary computation, a real-coded genetic algorithm, UNDX+MGG, shows good performance on multimodal functions with epistasis among parameters. However, UNDX+MGG has a problem that its performance is good on functions with big valley structures but deteriorates on those with the UV structures. On the other hand, ISM shows good performance on functions with the UV structures. However, ISM has two problems that (1) it fails in search when the region of the V valley including the optimum is very narrow and (2) its performance deteriorates on functions with big valley structures. In this paper, we propose a new evolutionary algorithm that aims at overcoming the problems of UNDX+MGG and ISM and examine its effectiveness through some experiments.