In this thesis I demonstrate a novel application of chaotic dynamics to evolutionary algorithms, specifically in population size management. Typical evolutionary algorithms require a population size to be set as a parameter, which remains constant throughout execution. I created a new algorithm that can vary the population size chaotically or periodically, and do a series of performance tests comparing static, periodic, and chaotic population control. The problems targeted in these tests are chosen from both continuous and discrete multi-dimensional domains. I find that both chaotic and static population control perform well in certain situations; my evidence suggests that periodic population control is rarely a good choice. I also present additional analysis on the effects of the population dynamics and how they relate to mean population size and variance in the performance results. |