Approximating the Pareto set
Loading...
Date
2001-10-16
Journal Title
Journal ISSN
Volume Title
Publisher
Universität Dortmund
Alternative Title(s)
concepts, diversity issues, and performance assessment
Abstract
This paper adresses the problem of diversity in multiobjective evolutionary algorithms and its implications for the quality of the approximated set of efficient solutions (Pareto set). Current approaches for maintaining diversity are classified and related to the overall fitness assignment strategy. The resulting groups of complex selection operators are presented and tested on different objective functions exhibiting different levels of difficulty. For the assessment of the algorithmic performance a quality measure based on the notion of dominance is applied that reflects gain of information produced by the algorithm. This allows an online and time-dependent evaluation in order to characterize the dynamic behavior of an algorithm.
This paper adresses the problem of diversity in multiobjective evolutionary algorithms and its implications for the quality of the approximated set of efficient solutions (Pareto set). Current approaches for maintaining diversity are classified and related to the overall fitness assignment strategy. The resulting groups of complex selection operators are presented and tested on different objective functions exhibiting different levels of difficulty. For the assessment of the algorithmic performance a quality measure based on the notion of dominance is applied that reflects gain of information produced by the algorithm. This allows an online and time-dependent evaluation in order to characterize the dynamic behavior of an algorithm.
This paper adresses the problem of diversity in multiobjective evolutionary algorithms and its implications for the quality of the approximated set of efficient solutions (Pareto set). Current approaches for maintaining diversity are classified and related to the overall fitness assignment strategy. The resulting groups of complex selection operators are presented and tested on different objective functions exhibiting different levels of difficulty. For the assessment of the algorithmic performance a quality measure based on the notion of dominance is applied that reflects gain of information produced by the algorithm. This allows an online and time-dependent evaluation in order to characterize the dynamic behavior of an algorithm.