Research on decomposition-based multi-objective evolutionary algorithm with dynamic weight vector

In recent years, multi-objective evolutionary algorithm based on decomposition has gradually attracted people's interest. However, this algorithm has some problems. For example, the diversity of the algorithm is poor, and the convergence and diversity of the algorithm are unbalanced. In additio...

Full description

Saved in:
Bibliographic Details
Published in:Journal of computational science Vol. 81; p. 102361
Main Authors: Zhao, Jiale, Huang, Xiangdang, Li, Tian, Yu, Huanhuan, Fei, Hansheng, Yang, Qiuling
Format: Journal Article
Language:English
Published: Elsevier B.V 01.09.2024
Subjects:
Combination evolution operator
Decomposition
Dynamic weight vector
Multi-objective evolutionary algorithm
Preference distribution
Combination evolution operator
Decomposition
Multi-objective evolutionary algorithm
Dynamic weight vector
Preference distribution
ISSN:1877-7503
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In recent years, multi-objective evolutionary algorithm based on decomposition has gradually attracted people's interest. However, this algorithm has some problems. For example, the diversity of the algorithm is poor, and the convergence and diversity of the algorithm are unbalanced. In addition, users don't always care about the entire Pareto front. Sometimes they may only be interested in specific areas of entire Pareto front. Based on the above problems, this paper proposes a decomposition-based multi-objective evolutionary algorithm with dynamic weight vector (MOEA/D-DWV). Firstly, a weight vector generation model with uniform distribution or preference distribution is proposed. Users can decide which type of weight vector to generate according to their own wishes. Then, two combination evolution operators are proposed to better balance the convergence and diversity of the algorithm. Finally, a dynamic adjustment strategy of weight vector is proposed. This strategy can adjust the distribution of weight vector adaptively according to the distribution of solutions in the objective space, so that the population can be uniformly distributed in the objective space as much as possible. MOEA/D-DWV algorithm is compared with 9 advanced multi-objective evolutionary algorithms. The comparison results show that MOEA/D-DWV algorithm is more competitive. Data will be made available on request. •A weight vector generation model with uniform distribution or preference distribution is proposed.•An improved Tchebycheff aggregation function is proposed to better guide population evolution.•Two combination evolution operators are proposed to better balance the convergence and diversity.•A dynamic adjustment strategy of weight vector is proposed.•In the design of the algorithm framework, Pareto dominance relation and external archiving are no longer used.
ISSN:1877-7503
DOI:10.1016/j.jocs.2024.102361