Self-adaptive weight vector adjustment strategy for decomposition-based multi-objective differential evolution algorithm

In multi-objective and many-objective optimization, weight vectors are particularly crucial to the performance of decomposition-based optimization algorithms. The uniform weight vectors are not suitable for complex Pareto fronts (PFs), so it is necessary to improve the distribution of weight vectors...

Full description

Saved in:
Bibliographic Details
Published in:Soft computing (Berlin, Germany) Germany), 2020-09, Vol.24 (17), p.13179-13195
Main Authors: Fan, Rui, Wei, Lixin, Li, Xin, Zhang, Jinlu, Fan, Zheng
Format: Article
Language:English
Subjects:
Algorithms
Artificial Intelligence
Computational Intelligence
Control
Convergence
Decomposition
Engineering
Evolutionary algorithms
Evolutionary computation
Mathematical Logic and Foundations
Mechatronics
Methodologies and Application
Multiple objective analysis
Optimization
Pareto optimization
Robotics
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:In multi-objective and many-objective optimization, weight vectors are particularly crucial to the performance of decomposition-based optimization algorithms. The uniform weight vectors are not suitable for complex Pareto fronts (PFs), so it is necessary to improve the distribution of weight vectors. Besides, the balance between convergence and diversity is a difficult issue as well in multi-objective optimization, and it becomes increasingly important with the augment of the number of objectives. To address these issues, a self-adaptive weight vector adjustment strategy for decomposition-based multi-objective differential evolution algorithm (AWDMODE) is proposed. In order to ensure that the guidance of weight vectors becomes accurate and effective, the adaptive adjustment strategy is introduced. This strategy distinguishes the shapes and adjusts weight vectors dynamically, which can ensure that the guidance of weight vectors becomes accurate and effective. In addition, a self-learning strategy is adopted to produce more non-dominated solutions and balance the convergence and diversity. The experimental results indicate that AWDMODE outperforms the compared algorithms on WFG suites test instances, and shows a great potential when handling the problems whose PFs are scaled with different ranges in each objective.
ISSN:1432-7643
1433-7479
DOI:10.1007/s00500-020-04732-y