Effects of migration methods in parallel distributed parameter‐free genetic algorithm
Susumu Adachi, Hidefumi Sawai · Electronics and Communications in Japan (Part II Electronics) · 2002
Abstract The effects of migration in parallel distributed processing of a parameter‐free genetic algorithm (PfGA) that does not require establishing genetic parameters such as the initial population size, crossing‐over rate, and mutation rate are discussed. This PfGA is a compact and robust algorithm characterized by the fact that it extracts one local population (subpopulation) from the entire search space and searches the prospective space while varying the local population size adaptively. The algorithm has been implemented on parallel computers using a hierarchical parallel‐distributed architecture of the PfGA that includes migration methods. As a result of performance evaluation using a general functional optimization (minimum value search) problem including the international benchmark problem, it has been verified that the search success rate is maintained or improved, and that increasing the number of local populations by using the proposed migration selection methods can effectively reduce the number of evaluations required for success. © 2002 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 85(11): 71–80, 2002; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/ecjb.10096