Adaptively Resizing Populations: Algorithm, Analysis, and First Results
Robert E. Smith
Department of Engineering Science and Mechanics,
The University of Alabama,
Box 870278, Tuscaloosa, Al 35487 USA.
Ellen Smuda
Department of Aerospace Engineering,
The University of Alabama,
Tuscaloosa, Al 35487 USA
Abstract
Deciding on an appropriate population size for a given genetic algorithm (GA) application can often be critical to the success of the algorithm. Too small, and the GA can fall victim to sampling error, affecting the efficacy of its search. Too large, and the GA wastes computational resources. Although advice exists for sizing GA populations, much of this advice involves theoretical aspects that are not accessible to the novice. This paper suggests an algorithm for adaptively resizing GA populations. The algorithm is suggested based on recent theoretical developments that relate population size to schema fitness variance. The algorithm is developed theoretically, simulated with expected value equations, and tested on a problem where population sizing can mislead the GA. The positive results presented suggest that adaptively sizing GA populations may be a practical extension to the typical GA. Such an extension frees the user from a critical parameter decision, and expands the usefulness of GA search. Moreover, this extension creates a new, interesting class of genetic search systems, where adaptive changes in population size reflect problem complexity.
