Alexis Ioannou Undergraduate Dissertation 2017/18

Adapting the population size in evolutionary algorithms

Supervised by D.Sudholt

Abstract

Evolutionary algorithms can be used to find optimal solutions to simple and difficult problems and it is inspired by biological evolutions. An evolutionary algorithm starts by having an initial population that produces next generations using techniques like mutations, crossover and selection. Furthermore, each generation has fitness values that are being calculated by an evaluation method and that determines how good the solutions are. Following this, the best results of the generation and the new offsprings are passed on to the next generation. The results can vary depending on what parameters are given to the EA ( population size, mutation probability, crossover probability). The goal of this project is to implement different schemes that adapt dynamically the offspring population size while the algorithm runs and test and compare their performance across a range of problems.