Learning-Driven Algorithm With Dual Evolution Patterns for Solving Large-Scale Multiobjective Optimization Problems

Recently, a series of evolutionary algorithms have been proposed to enhance the search efficiency when handling large-scale multiobjective optimization problems (LSMOPs). Among them, learning-driven evolutionary algorithms have shown to be promising by learning the related evolutionary pattern with...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access Vol. 13; pp. 30976 - 30992
Main Authors: Song, Mingshuo, Song, Wei, Wee Lai, Khin
Format: Journal Article
Language:English
Published: Piscataway IEEE 2025
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Subjects:
Convergence
differential evolution
Evolutionary algorithms
Evolutionary computation
evolutionary pattern learning
Focusing
Genetic algorithms
Indexes
Large-scale multiobjective optimization
learning-aided evolution
Machine learning
Marine vehicles
Multiple objective analysis
neural network
Neural networks
Optimization
Search problems
Training
Vectors
ISSN:2169-3536, 2169-3536
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Recently, a series of evolutionary algorithms have been proposed to enhance the search efficiency when handling large-scale multiobjective optimization problems (LSMOPs). Among them, learning-driven evolutionary algorithms have shown to be promising by learning the related evolutionary pattern with a specific neural network. However, the learned evolutionary patterns primarily focus on convergence while ignoring diversity, which may potentially cause an inadequate search. In this paper, we propose a learning-driven algorithm with dual evolution patterns (DEPLA) for solving LSMOPs. The main idea is to learn two evolution patterns in the evolutionary process, one focusing on convergence and the other on diversity, which would be helpful to produce promising solutions efficiently. To be specific, DEPLA first divides the population into two parts to be updated with different strategies. A half of the solutions with good convergence and distribution will be updated with the learned evolution pattern focusing on convergence, while the rest of the solutions are repositioned by another learned evolution pattern towards diversity. Through learning the dual evolution patterns in both convergence and diversity, DEPLA is capable of efficiently tackling LSMOPs. Abundant experiments are conducted on the well-known LSMOPs test suites with up to 10,000 decision variables. The experimental results demonstrate that our DEPLA outperforms six state-of-the-art large-scale multiobjective evolutionary algorithms in handling LSMOPs.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2025.3541271