10th EAI International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS)

Research Article

Affinity Based Search Amount Control in Decomposition Based Evolutionary Multi-Objective Optimization

Download46 downloads
  • @INPROCEEDINGS{10.4108/eai.22-3-2017.152400,
        author={Hiroyuki Sato and Minami Miyakawa and Keiki Takadama},
        title={Affinity Based Search Amount Control in Decomposition Based Evolutionary Multi-Objective Optimization},
        proceedings={10th EAI International Conference on Bio-inspired Information and Communications Technologies (formerly BIONETICS)},
        publisher={EAI},
        proceedings_a={BICT},
        year={2017},
        month={3},
        keywords={multi-objective optimization evolutionary algorithms moea/d},
        doi={10.4108/eai.22-3-2017.152400}
    }
    
  • Hiroyuki Sato
    Minami Miyakawa
    Keiki Takadama
    Year: 2017
    Affinity Based Search Amount Control in Decomposition Based Evolutionary Multi-Objective Optimization
    BICT
    EAI
    DOI: 10.4108/eai.22-3-2017.152400
Hiroyuki Sato1,*, Minami Miyakawa2, Keiki Takadama1
  • 1: The University of Electro-Communications
  • 2: Hosei University, JSPS Research Fellow (PD)
*Contact email: h.sato@uec.ac.jp

Abstract

This work proposes a search amount control method on each search part of the Pareto front in decomposition based evolutionary multi-objective optimization. The conventional MOEA/DC decomposes the Pareto front with a set of weight vectors and pairs one solution with each weight vector to approximate the entire Pareto front with the set of solutions. Well-matched pairs of weight vector and solution contribute to uniformly approximating the Pareto front, and mismatched pairs having a long distance between weight vector and solution in the objective space deteriorate the approximation quality and the search. To eliminate mismatched pairs and improve the search performance, this work proposes affinity based search amount control method for MOEA/DC. Experimental results using continuous WFG4 test problems with 2-5 objectives show that the proposed method improves the well-matched pair ratio in all pairs of weight vector and solution and the search performance.