Finding informative collaborators for cooperative co-evolutionary algorithms using a dynamic multi-population framework

Cooperative co-evolutionary algorithms (CCEAs) conduct high-efficiency problem solving by decomposing a given problem into a number of separate subcomponents, which terms the divide-and-conquer manner. In this paper, the dynamic multi-population framework was incorporated into the CCEAs to continuously search multiple optima of the subcomponents, so as to compensate the lost information induced by problem decomposition and enhance the global optimization capability. These optima are seen as the informative collaborators that can feature the landscapes of the subcomponents. Thus, more accurate fitness evaluation could be conducted by mixing those collaborators. To verify this idea, two dynamic multi- population optimizers were implemented, which results in two dynamic multi-population based CCEAs. Experimental study was carried out on a wide range of benchmark functions. The proposed algorithms was compared with four peer algorithms to verify the effectiveness.