A Relation-Based Model for Convergence Analysis of Evolutionary Algorithm

摘要

There have been many results on convergence of evolutionary algorithm (EA) since it was proposed, but few result focused on convergence analysis based on relation theory. This paper proposed a relation-based model to study the equivalence and ordering of EA in convergence. The equivalence relation named equivalence in status (EIS) can be used to divide a given set of EAs into equivalence classes in which the EAs have the same capacity of convergence. EAs belonging to different EIS classes have different capacities of convergence based on the absorbing Markov chain model, which is described as an ordering relation named superiority in status. The performance of an EA can be improved if it is modified to be superior in status to its original version.