Make fast evolutionary programming robust by search step control

Yong LIU, Xin YAO

Research output: Book Chapters | Papers in Conference ProceedingsConference paper (refereed)Researchpeer-review

Abstract

It has been found that both evolutionary programming (EP) and fast EP (FEP) could get stuck in local optima on some test functions. Although a number of methods have been developed to solve this problem, nearly all have focused on how to adjust search step sizes. This paper shows that it is not enough to change the step sizes alone. Besides step control, the shape of search space should be changed so that the search could be driven to other unexplored regions without getting stuck in the local optima. A two-level FEP with deletion is proposed in this paper to make FEP robust on finding better solutions in function optimisation. A coarse-grained search in the upper level could lead FEP to generate a diverse population, while a fine-grained search in the lower level would help FEP quickly find a local optimum in a region. After FEP could not make any progress after falling in a local optimum, deletion would be applied to change the search space so that FEP could start a new fine-grained search from the points generated by the coarse-grained search. © Springer-Verlag Berlin Heidelberg 2006.
Original languageEnglish
Title of host publicationAdvances in Natural Computation : Second International Conference, ICNC 2006, Xi'an, China, September 24-28, 2006, Proceedings, Part I
EditorsLicheng JIAO, Lipo WANG, Xin-bo GAO, Jing LIU, Feng WU
PublisherSpringer Berlin Heidelberg
Pages806-815
Number of pages10
ISBN (Electronic)9783540459026
ISBN (Print)9783540459019
DOIs
Publication statusPublished - 2006
Externally publishedYes
Event2nd International Conference on Natural Computation, ICNC 2006 - Xi'an, China
Duration: 24 Sept 200628 Sept 2006

Publication series

NameLecture Notes in Computer Science
PublisherSpringer Berlin, Heidelberg
Volume4221
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference2nd International Conference on Natural Computation, ICNC 2006
Country/TerritoryChina
CityXi'an
Period24/09/0628/09/06

Fingerprint

Dive into the research topics of 'Make fast evolutionary programming robust by search step control'. Together they form a unique fingerprint.

Cite this