Optimal Generational Model

The type of generational model in EAs determines whether or not best solutions found so far should be kept for the next generation. In the overlapping model, the best solution(s) from the parent population can compete for survival with solutions from the offspring population. On the other hand, in the nonoverlapping model only solutions from the offspring population can compete for survival.

In order to determine the impact of the type of the generational model (overlapping vs. nonoverlapping) on MAs, a series of design experiments involving two kinds of MA-ES was conducted, namely MA-ES(5+25) (overlapping) and MA-ES(5,25) (nonoverlapping). The experiments included a total of 24 design experiments (12 for each algorithm) utilizing all 12 combinations of mutation and crossover rates (see Table 3). Figure 12 shows typical results obtained in these experiments. Here, mutation and uniform crossover rates were equal to 0.025 and 0.2, respectively. Figure 12 shows that there are no significant differences between MA-ES(5,25) and MA – ES(5+25). This type of behavior was observed in all conducted experiments. In several cases MA – ES(5+25) slightly outperformed MA-ES(5,25) but in other cases it produced inferior results. The differences between the two generational models were, however, small both in terms of variance and fitness of the generated designs. Generally, it can be concluded that MA-ES with the overlapping and nonoverlapping generational model produce comparable results in these problem domains.