基于MPICH平台的多种群并行遗传算法

提出了一种基于集群系统的多种群并行遗传算法，在集群系统的软件开发环境MPICH上构建了多种群粗粒度并行遗传算法框架．采用全局迁移方式交换通信域中各子群体的个体信息，并通过模拟退火规则来确定迁移代频．利用该算法框架对一类优化问题——N维目标函数的最小值问题，给出了具体的实现方法．最后对该类问题的两个实例进行了测试，基于该实例的测试数据对论文提出的算法进行了性能分析．结果显示，算法具有线性加速比，而且解的精度随着并行进程数量的增多而提高．

A Multiple-population Parallel Genetic Algorithm Based on Platform MPICH

Based on the cluster system's software development environment MPICH,a multiple-population parallel genetic algorithm framework is presented.In this algorithm framework,every sub-population can exchange information of its individuals with all the others through global migrating among the sub-populations in the communicating community of the cluster,and each sub-population determines its migrating frequency by the rule of simulating anneal.This paper also shows a solution to a kind of optimization problems which seek the minimal value of the N-Dimension function using the algorithm framework.Finally,two examples of the mentioned problems are tested and the performance of the multiple-population parallel genetic algorithm is analyzed by the test data.The result reveals that the algorithm not only has a linear speedup ratio but also can obtain a more optimal value as the number of processes of the communicating community increasing.