Controlled markov chain optimization of genetic algorithms

Document Type

Conference Proceeding

Publication Date

1-1-1999

Abstract

Identifying the optimal settings for crossover probability pc and mutation probability Pm is of an important problem to improve the convergence performance of GAs. In this paper, we modelled genetic algorithms as controlled Markov chain processes, whose transition depend on control parameters (probabilities of crossover and mutation). A stochastic optimization problem is formed based on the performance index of populations during the genetic search, in order to find the optimal values of control parameters so that the performance index is maximized. We have shown theoretically the existence of the optimal control parameters in genetic search and proved that, for the stochastic optimization problem, there exists a pure deterministic strategy which is at least as good as any other pure or mixed (randomized) strategy.

Identifier

84956648021 (Scopus)

ISBN

[354066050X, 9783540660507]

Publication Title

Lecture Notes in Computer Science Including Subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics

External Full Text Location

https://doi.org/10.1007/3-540-48774-3_22

e-ISSN

16113349

ISSN

03029743

First Page

186

Last Page

196

Volume

1625

This document is currently not available here.

Share

COinS