The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #94123

Title:Analysis of Island Ga
Date: 1994
Abstract:Island GA is a kind of parallel GA, distinguished from traditional or canonical GA, where a whole population is subdivided into smaller subpopulations. Extensive experiments are conducted for two versions of island GA on De Jong's third problem and Royal Road problem. The two versions are discrete island GA and tagged island GA. It is found that for discrete island GA, with suitable control of parameters: number of subpopulation, migration interval, migration rate, migration method, its performance can be better than canonical GA. For tagged island GA, its performance is similar to discrete island GA without migration. Knowing that discrete island GA can have better performance, analysis is concentrated on it to identify the reasons for the improved performance. Three factors are identified: 1. effect of getting global optimum from crossover of sub-optima; 2. island effect; 3. migration effect.

[Search These Pages] [DAI Home Page] [Comment]