Cooperative coevolutionary algorithms for dynamic optimization: an experimental study

In this paper, we study the cooperative coevolutionary algorithms (CCEAs) for dynamic optimization. We introduce the CCEAs with two popular types of individuals: (1) random immigrants (RIs) that increase the diversity for changing environments, and (2) elitist individuals that increase the local convergence to the optima. The CCEAs are evaluated on a standard suite of benchmark problems and are compared with evolution strategies (ES). Our experimental results show that the CCEAs are efficient in locating and tracking optima in dynamic environments. They are superior to the ES when the RI individuals and the elitist individuals are used. In addition, we empirically investigate how the CCEAs perform with different parameter settings. These settings include collaboration methods, the use of plus- comma selections, and the number of RI individuals and elitist individuals. We also investigate the CCEAs that use a mutative sigma-self adaptation. The CCEAs perform the best when they use the best collaboration method and the plus selection. The use of the mutative sigma-self adaptation is insignificant. Our results also show that the CCEAs are more scalable than the ES in dynamic environments.