Empirical study of effect of grouping strategies for large scale optimization

The cooperative co-evolution framework (CC) is widely used in the large scale global optimization. It is believed that the CC framework is very sensitive to grouping strategies and the performance deteriorate if interacted variables are not correctly grouped. So many efforts have been devoted to find good ways to correctly decompose the large scale problem into smaller sub-problems so as to effectively solve the original problem by optimizing these smaller sub-problems using a search algorithm. However, what is the relationship between the grouping strategy and the search algorithm adopted in CC? what is the effect of grouping strategies on the CC framework? This work will tackle these issues. We try to unveil the impact of different grouping strategies on CC and the relationship between the grouping strategies and the search algorithms by empirical study. The experiment results show that the correct result of variable grouping is very important since it can turn the large scale problem into smaller sub-problems and make the problem solving easier. It indeed has a big influence on the results obtained by the search algorithm. However, when the search algorithm adopted is not suitable or effective, even if the grouping strategy gives the correct grouping results, the final results may be poor. In this case, grouping strategy only plays little role on the CC. Thus, only effective grouping strategy plus efficient search algorithm can result in good solutions for large global optimization problems.