An improved global replacement strategy for MOEA/D on many-objective kanpsack problems
Xingxing Hao, Jing Liu, Zhenkun Wang · 2017
The multi-objective evolutionary algorithm based on decomposition (MOEA/D) decomposes a multi-objective optimization problem into a number of single scalar optimization problems and solves them simultaneously. The replacement strategy employed in MOEA/D has significant effects in terms of balancing convergence and diversity. In this paper, the effectiveness of MOEA/D with global replacement (GR) scheme is first investigated on many-objective knapsack problems. Then, we propose an improved version of GR, which is denoted as IGR, for the situation of adopting the utopian point as the reference point in MOEA/D. The experimental results on knapsack problems with 2, 4, 6, and 8 objectives illustrate that the GR scheme outperforms the original MOEA/D adopting the ideal point as the reference point and the IGR scheme outperforms the original MOEA/D adopting the utopian point as the reference point.