differential evolution, optimization, parallel models

Two types of coarse-grain parallel migration model for EAs are studied. Models are applied to Success-History Based Parameter Adaptation for Differential Evolution (SHADE) algorithm. The population is divided into set of islands in which independent search for global minimim point is executed. Individual exchanges, migrations are performed repeatedly among islands after certain number of generations. Two types of migration strategies are employed, namely star topology and ring topology. Different settings of parallel models, e.g. number of islands and migrations are tested on benchmark suite. The results show that performance of proposed algorithms highly depends on these settings. Parallel speedup of algorithms is compared to sequential version of SHADE.