Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
A MOEA/D Algorithm of Weight Vectors and Neighborhood Size with Adaptive Learning
1School of Information Engineering, JiangXi University of Science and Technology, Ganzhou, China.
2School of Mathematical and Computer Science, Gannan University, Ganzhou, China.
ABSTRACT
The multiobjective evolutionary algorithm based on decomposition (MOEA/D) is an effective method for solving multiobjective optimization problems. In classical MOEA/D, the fixed weight vectors allocates the same computing resources to all regions, and the neighborhood size remains constant during the evolution of the algorithm. These characteristics waste computing resources and affect the performance of the algorithm when dealing with multiobjective optimization problems that have a discontinuous Pareto front (PF). To solve the above problems, a MOEA/D algorithm of weight vectors and neighborhood size with adaptive learning (MOEADWT) is proposed in this paper. During the algorithm evolution process, this method deletes a certain number of weight vectors in crowded regions, generates new weight vectors based on reference points and weight vectors in sparse regions, and adaptively learns the neighborhood size based on the number of evaluations. In the experimental study, the proposed MOEADWT algorithm is compared with the other four algorithms. The experimental results show that the proposed algorithm is more suitable for dealing with discontinuous PF optimization problems, and the convergence and diversity of this algorithm are better than those of other algorithms.
Keywords: MOEA/D, Pareto front, Weight vector, Neighborhood size, Adaptive learning.
Download PDF
1School of Information Engineering, JiangXi University of Science and Technology, Ganzhou, China.
2School of Mathematical and Computer Science, Gannan University, Ganzhou, China.
ABSTRACT
The multiobjective evolutionary algorithm based on decomposition (MOEA/D) is an effective method for solving multiobjective optimization problems. In classical MOEA/D, the fixed weight vectors allocates the same computing resources to all regions, and the neighborhood size remains constant during the evolution of the algorithm. These characteristics waste computing resources and affect the performance of the algorithm when dealing with multiobjective optimization problems that have a discontinuous Pareto front (PF). To solve the above problems, a MOEA/D algorithm of weight vectors and neighborhood size with adaptive learning (MOEADWT) is proposed in this paper. During the algorithm evolution process, this method deletes a certain number of weight vectors in crowded regions, generates new weight vectors based on reference points and weight vectors in sparse regions, and adaptively learns the neighborhood size based on the number of evaluations. In the experimental study, the proposed MOEADWT algorithm is compared with the other four algorithms. The experimental results show that the proposed algorithm is more suitable for dealing with discontinuous PF optimization problems, and the convergence and diversity of this algorithm are better than those of other algorithms.
Keywords: MOEA/D, Pareto front, Weight vector, Neighborhood size, Adaptive learning.
Download PDF