Estimating the reproductive potential of offspring in evolutionary heuristics for combinatorial optimization problems
Date
2011
Authors
Parkinson, E.
Ghandar, A.
Michalewicz, Z.
Tuson, A.
Editors
Advisors
Journal Title
Journal ISSN
Volume Title
Type:
Conference paper
Citation
Proceedings of the 2011 IEEE Congress of Evolutionary Computation (CEC 2011), held in New Orleans, LA, USA, 5-8 June, 2011: pp.172-178
Statement of Responsibility
Eddy Parkinson, Adam Ghandar, Zbigniew Michalewicz and Andrew Tuson
Conference Name
IEEE Congress of Evolutionary Computation (2011 : New Orleans, USA)
Abstract
This paper proposes a metaheuristic selection technique for controlling the progress of an evolutionary algorithm (and possibly other heuristic search techniques) to manipulate and make use of the relationship between runtime and solution quality. The paper examines the idea that very rapid increases in initial fitness may lead to premature convergence and a reported solution that is less than optimal. We examine the advantages provided by this metaheuristic selection technique in solving two different combinatorial optimization problems: including a "toy" problem of finding magic squares and a more realistic vehicle routing problem (VRP) benchmark. The method is found to be useful for finding both higher quality solutions with a marginally longer algorithm run time and for obtaining lower quality solutions in a shorter time. Furthermore, the impact on the search results is similar for both the magic square and the VRP problem providing evidence the method is scalable to other problem domains, and therefore is potentially a relatively straight forward addition to many heuristic approaches that can add value by improving both runtime and solution quality.
School/Discipline
Dissertation Note
Provenance
Description
Access Status
Rights
©2011 IEEE