Cooperative co-evolution with graph-based differential grouping for large scale global optimization

Abstract

Cooperative co-evolution framework has been developed for solving large scale global optimization problems. This approach applies the divide-and-conquer strategy that decomposes the problem into subcomponents which can be optimized separately. Nevertheless, the decomposition strategy has important influence on solution quality. In theory, the interdependency between subcomponents should be kept minimum as the subcomponents coadaptation is needed to solve large scale optimization problems. Some state of the art decomposition strategies like differential grouping gain high grouping accuracy on a suite of benchmark functions. In this paper, we use graph theory to model the decomposition problem and propose graph-based differential grouping decomposition strategy to improve the decomposition accuracy of differential grouping. Empirical studies show that our decomposition method get perfect performance on all benchmark functions. Significantly, the solution quality on large scale problems are better than several outstanding decomposition strategies when the graph-based differential grouping is embedded with cooperative co-evolution framework.