Abstract: | A commonly encountered feature of genetic algorithms is that they tend to offer just one candidate solution to a problem. This project investigates methods to increase the number of solutions generated by genetic algorithms in the domain of simple timetabling problems.
Spatial selection [Collins and Jefferson 91], islands [Tanese 89], crowding [De Jong 75], sharing [Deb and Goldberg 89, Goldberg and Richardson 87], some simple combinations of these methods and a new variant of spatial selection called "tribes" were tested on four basic timetabling problems.
For problems with large numbers of solutions a simple GA run many times was found to produce more distinct solutions in a faster time than any of the specialised methods. On problems with low numbers of solutions spatial selection and tribes were found to be the most efficient at finding distinct solutions.
|