A New Strategy of Dynamically Adjusting Population Size for Coevolutionary Algorithms

摘要

How to dynamically adjust population size is one of the most important research topics in coevolutionary computation. This paper proposes a new strategy to dynamically adjust population size during a coevolutionary process. Three factors are considered in the strategy to describe the influence on population size. They are the natural growth of sub-populations, the internal competition of individuals from same sub-population and the interaction between individuals from different sub-populations. It is proven that the proposed strategy is globally asymptotically stable and also the convergence of a coevolutionary algorithm with the strategy is investigated. Finally, the behavior of the proposed strategy in practical use is illustrated.