Transmission Function Models of Finite Population Genetic Algorithms. Software211 Engineering

摘要

Infinite population models show a deterministic behavior. Genetic algorithms with211u001efinite populations behave non-deterministicly. For small population sizes, the 211u001eresults obtained with these models differ strongly from the results predicted by 211u001ethe infinite population model. When the population size is increased towards 211u001einfinity, a convergence to the results predicted by the infinite population 211u001emodels is observed. In real GA's random decisions are used during the run of the 211u001eGA. These random decisions can lead to a behavior that results in a deviation of 211u001ethe GA from the expected path of evolution. In this report four sources of non-211u001edeterminism are identified. Finite population models are generated by explicitly 211u001emodeling two of these sources. When comparing the results to runs of actual 211u001egenetic algorithms, similar results are obtained. Hence, this model shows what 211u001eare the most important sources of non-determinism in the GA for the problem at 211u001ehand.