Initial Population of NSGA-II for Solving Similar Multi-Objective Optimization Problems

When optimizing a highly complex real-world problem, the solution may not be sufficiently searched for, or it may take a long time to converge on the solution. On the other hand, companies often develop products with similar trends, such as improvements to existing products or product series. Therefore, when developing similar products using evolutionary multi-objective optimization, it is expected that solution search performance will improve by saving the design data of past products and including them in the initial solution. In this paper, we use NSGA-II to investigate the effect on the convergence speed to the Pareto front and the diversity of the solution distribution when a part of the non-dominated solution obtained in one problem is memorized and included in the initial solution when solving another similar problem. As a result, we show that the convergence speed improves when using solutions near the center of the Pareto front, which are commonly selected in product development, while solution search performance deteriorates for some benchmark problems, and that using solutions on the edge of the Pareto front is effective in improving the diversity of the solution distribution.