Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
A Multi-Objective Evolutionary Algorithm Based on Reduced-Dimensional Space Partition
1School of Computer Science and Technology, Jiangsu Normal University, Xuzhou, China.
2Jiangsu Wisdom-driven Research Institute Co., Ltd, Xuzhou, China.
ABSTRACT
The Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is well-known to solve multiobjective optimization problems. However, due to insufficient acquisition and ineffective utilization of Pareto Front (PF) information, the MOEA/D does not work well on problems with discontinuous PF. Therefore, we propose a MOEA/D based on Reduced-Dimensional Space Partition (MOEA/D-RDSP) by using clustering algorithm and partitioning method. MOEA/D-RDSP contains two parts and performs global MOEA/D in the first part. In the second part, the clustering algorithm is adopted to perceive the distribution of discontinuous PF and the partitioning method is adopted to uniformly divide the objective space. The partition result is used to update the neighborhood of each weight vector, and local MOEA/D is adopted to update the external population of each subspace, which is followed by the update of PF. The algorithm is validated on 10 classical multi-objective optimization benchmarks. The results show that our method outperforms the other 4 popular algorithms.
Keywords: MOPs, MOEA/D, Clustering algorithm, Objective space partition, Dimension reduction, Pareto front.
Download PDF
1School of Computer Science and Technology, Jiangsu Normal University, Xuzhou, China.
2Jiangsu Wisdom-driven Research Institute Co., Ltd, Xuzhou, China.
ABSTRACT
The Multi-Objective Evolutionary Algorithm based on Decomposition (MOEA/D) is well-known to solve multiobjective optimization problems. However, due to insufficient acquisition and ineffective utilization of Pareto Front (PF) information, the MOEA/D does not work well on problems with discontinuous PF. Therefore, we propose a MOEA/D based on Reduced-Dimensional Space Partition (MOEA/D-RDSP) by using clustering algorithm and partitioning method. MOEA/D-RDSP contains two parts and performs global MOEA/D in the first part. In the second part, the clustering algorithm is adopted to perceive the distribution of discontinuous PF and the partitioning method is adopted to uniformly divide the objective space. The partition result is used to update the neighborhood of each weight vector, and local MOEA/D is adopted to update the external population of each subspace, which is followed by the update of PF. The algorithm is validated on 10 classical multi-objective optimization benchmarks. The results show that our method outperforms the other 4 popular algorithms.
Keywords: MOPs, MOEA/D, Clustering algorithm, Objective space partition, Dimension reduction, Pareto front.
Download PDF