The Impact of Population Size on the Convergence of Multi-objective Evolutionary Algorithms

In this paper, we evaluate the effects of different population sizes on the performance and convergence behavior of the initial population in multi-objective evolutionary algorithms (MOEAs). While previous studies on the population size often concern the trade-off between runtime and performance, we provide an additional focus on the changes in convergence behavior of the initial population. For this, the gene tracking approach of the traceable evolutionary algorithm (T-EA) is used with multi-objective problems (MOPs) as the traceable multi-objective evolutionary algorithm (T-MOEA). Through this approach, the influence of the individuals from the initial population on the final generation can be evaluated for MOEAs. Furthermore, the concept of contributing individuals is proposed as an additional metric for evaluation. To evaluate the influence of different population sizes, nine benchmarking problems are run using five different population sizes. Both the performance of the test problems and the convergence of the initial population are discussed in the evaluation.