A New Selection Strategy for Decomposition-based Evolutionary Many-Objective Optimization

It is challenging for optimization algorithms to obtain well converged and diverse optimal solutions in solving many-objective optimization problems, especially when the Pareto front is complex. In this paper, a new selection strategy is proposed for decomposition based evolutionary algorithms for solving many-objective optimization problems. In the proposed method, each individual in a population is assigned to a reference vector at first according to the angle between the objective vector of this individual and the reference vectors to divide the population into some subpopulations. Then for each subpopulation, an ideal point will be specified according to the minimum value of each objective among all solutions in the corresponding subpopulation. The individual in each subpopulation with the maximum ratio of the cosine of the angle between its objective vector and the corresponding reference vector to the distance from the individual to the corresponding ideal point will be selected to be passed to the next generation. The proposed algorithm is compared with five state-of-the-art algorithms on the DTLZ1-7 and WFG3-4 test problems with up to 15 objectives. The experimental results showed the competitiveness of the proposed method.