Replacement Strategies in Steady-State Multi-objective Evolutionary Algorithm: A Comparative Case Study

The paper investigates different replacement strategies for an elite, steady-state multi-objective evolutionary algorithm (MOEA), where the main population and archive population are evolved simultaneously. A new archive population replacement strategy is presented by updating the archive population with offspring based on epsilon-dominance relationship, which is able to maintain the extreme individuals and obtain a subset of Pareto optimal set without degradation. Furthermore, six different replacement strategies are implemented on the main population, which leads to different interaction between the main population and the archive population. All strategies are tested on two kinds of scalable test problems, and experimental results illustrate that high elitism is preferred in the steady-state MOEAs. Moreover, boundary points can effectively expand the final Pareto front in objective space when they are used in genetic operations.