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
Recommended Citation
Cao, Yijia and Cao, Lilian, "Controlled markov chain optimization of genetic algorithms" (1999). Faculty Publications. 16113.
https://digitalcommons.njit.edu/fac_pubs/16113
