An improved MOEA/D design for many-objective optimization problems

摘要

MOEA/D is one of the most popular multi-objective evolutionary algorithms. To extend the effective application scope of MOEA/D for high-dimensional objectives, an improved MOEA/D design for many-objective optimization problems, named I-MOEA/D, is proposed in this paper. Comparing with the original MOEA/D, we redesigned the weight vectors used in the subproblems, making the distribution of the weight vectors broader and more effective to ensure the diversity and convergence of solutions in the objective space. Moreover, a new decomposition approach, called the weighted mixture-style method, which combines the advantages of the weighted sum decomposition and the Tchebycheff decomposition approaches, is adopted in I-MOEA/D to improve the effectiveness of the algorithm. A three-part experimental comparison using DTLZ1-DTLZ4, with the number of objectives ranging from three to fifteen, is performed. Experimental results verify the effectiveness of each strategy and reveal that the proposed I-MOEA/D method achieves better performance than the other related state-of-the-art algorithms in solving this type of many-objective optimization problems.