Parallel genetic algorithm based on unoriented-connected topology and the improvement of mutation operator

Xiang-yong XU,Hong-cheng YIN,Chun-lei LI,Xiao WEI

DOI: https://doi.org/10.3321/j.issn:1001-506X.2005.05.037

2005-01-01

Abstract:In order to solve the massive optimization problems, a novel topology-unoriented-connected topology is presented based on Internet for parallel genetic algorithm. Some computers in which the subpopulations reside are allowed to quit in the midway without any limitation, and some other computers are allowed to participate in the evolution, such that the robustness and fault-tolerance of the algorithm are enhanced. At the same time, according to the disadvantages of the traditional floating-point mutation operator, a dyadic floating-point supplementary mutation operator is brought forward. The function of the novel mutation operator to prevent premature convergence is also discussed. Experimental results show that parallel genetic algorithm based on unoriented-connected topology can not only gain better optimization results but also largely save the optimization time of the problems. The mutation operator can effectively prevent genetic algorithm from being trapping in local extremum, and the searching performance of genetic algorithm is largely improved.