Diversity Guided Adaptive Evolutionary Algorithm

Parameter tuning in Evolutionary Algorithms (EA) generally result in suboptimal choices of values because of the complex dependencies between the parameters. Furthermore, different scenarios during a run of the EA often have different optimal parameter values. This thesis aims to better the understanding of how information about previously successful applications of genetic operators can be used to improve the quality of the search by using derandomised self-adaptive parameter control; We utilise the genetic differences between an offspring its parent to adapt a mutation vector. It also explores two different selection strategies that maintains diversity in the population, and the general effect that diversity has on the exploration and exploitation of the solution space. The adaptive mutation scheme proposed in this thesis has shown to improve the speed of the EA significantly while still being able to solve a wide range of mathematical functions as well as practical problems. Supplemented with a simple scheme that maintains diversity it becomes a more robust implementation well suited for multiple types of problems; especially for problems with computationally expensive fitness tests