The automatic design of parameter adaptation techniques for differential evolution with genetic programming

This study proposes a technique aimed at the automatic search for parameter adaptation strategies in a differential evolution algorithm with genetic programming symbolic regression. Genetic programming is applied to find the symbolic expression for scaling factor control during the optimization proc...

Full description

Saved in:
Bibliographic Details
Published in:Knowledge-based systems Vol. 239; p. 108070
Main Authors: Stanovov, Vladimir, Akhmedova, Shakhnaz, Semenkin, Eugene
Format: Journal Article
Language:English
Published: Amsterdam Elsevier B.V 05.03.2022
Elsevier Science Ltd
Subjects:
Adaptation
Algorithms
Automatic
Design factors
Design parameters
Differential evolution
Evolution
Evolutionary algorithms
Evolutionary computation
Extraction
Genetic algorithms
Genetic programming
Genetics
Information sources
Optimization
Parameter adaptation
Programming
Property
Scaling factors
Symbolism
Symbols
Differential evolution
Parameter adaptation
Genetic programming
ISSN:0950-7051, 1872-7409
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:This study proposes a technique aimed at the automatic search for parameter adaptation strategies in a differential evolution algorithm with genetic programming symbolic regression. Genetic programming is applied to find the symbolic expression for scaling factor control during the optimization process of differential evolution based on the current computational resource, ratio of successful solutions and adapted scaling factor value. The design of the parameter adaptation technique is performed by a computational experiment, which consisted in solving several complex optimization problems. Better symbolic expressions are selected with regards to the Friedman ranking procedure, and the best solutions are additionally evaluated to compare them to the existing parameter adaptation techniques. The experimental results show that the automatically designed parameter adaptation techniques described by symbolic expressions are capable of outperforming existing parameter adaptation methods, while using different information sources. The analysis of automatically generated solutions shows that the proposed technique can be considered an automatic knowledge extraction method. This is due to the results showing that well-performing parameter adaptation can behave differently from state-of-the-art methods, thereby revealing previously unknown algorithm properties.
Bibliography:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0950-7051
1872-7409
DOI:10.1016/j.knosys.2021.108070