INVESTIGATION OF THE IMPACT OF POPULATION SIZE ON THE PERFORMANCE OF A GENETIC ALGORITHM

Abstract

The paper investigates ways to determine the population size in a genetic algorithm and studies the
relationship between the number of individuals and the speed of the algorithm. Methods for determining
the optimal number of individuals in a population by different methods are described: depending on the
size of the chromosomes, for a tree-like type of chromosomes, in the presence of a noise factor and by the
method of a neighboring element with a maximum and minimum boundary. The data obtained by performing
each method differ from each other, for this reason, an assessment was made in order to verify the
accuracy of theoretical data by comparing them with experimental ones. To conduct experiments, a program was developed on the Unity graphics platform with the ability to change the number of individuals in
the population. After receiving the results, the experimental data were compared with the data obtained on
the basis of methods for determining the population size in the genetic algorithm from the first part of the
work. The experiment showed that the optimal population size lies in the range of 100-160 individuals.
With a decrease in their number, the execution time of the task begins to increase significantly, and with
an increase beyond the calculated limit, the reduction in execution time does not correspond to the computing
resources expended. The experimental data obtained themselves have the smallest error with the
method used by the tree representation of chromosomes. The results of the study can be used to select the
size of the population during training in order to achieve a better ratio of computing power to learning
speed, and a method defined in the course of work can help in theoretical calculations