Date of Award
Master of Science
Graph based evolutionary algorithms (GBEAs) have been used to control the rate of information flow and increase diversity in evolutionary algorithms, but there is limited understanding of the mechanics of this spread of information. This thesis proposes a metric, k[greek letter kappa], to quantify the rate at which information spread across an evolving graph based population. This metric provides a correlation between population size, graph type, and time for a good solution to spread across the entire graph. k[greek letter kappa] is roughly equivalent to the graph diameter for the graphs study. This metric can be used to compare population structures and sizes. This metric is then used to investigate the impact of the population size on the performance of GBEAs. As population size grows a trade off exists between preserving diversity and increased information spread. In most of the problems studied, there was an critical population size. In populations smaller than the critical population increased diversity, e.g. more disconnected graph based populations, resulted in faster solutions. In evolving populations larger than the critical population increased rate of information spread, e.g. more tightly connected graph based populations, resulted in faster solutions. Using these observations provides an initial basis for developing a method for determining a priori which graph would best preserve diversity, allowing for more difficult problems to be solved and/or producing multiple solutions to engineering design problems.
Steven Michael Corns
Corns, Steven Michael, "Measuring information exchange rate for graph based evolutionary algorithms in engineering optimization" (2003). Retrospective Theses and Dissertations. 19935.