一种粗粒度并行遗传算法及其应用  

A parallel genetic design method with coarse grain

作　　者：武金瑛[1] 王希诚[1] 

机构地区：[1]大连理工大学工业装备结构分析国家重点实验室,大连116024

出　　处：《计算力学学报》2002年第2期148-153,共6页Chinese Journal of Computational Mechanics

基　　金：国家自然科学基金 (5 97780 44 );国家高性能计算基金 (9910 1);国家 973项目 (G19990 3 2 80 5 )资助 .

摘　　要：提出一种粗粒度并行遗传算法 ,采用多种群遗传策略 ,在逐渐收缩的空间上搜索 ,并且用准精确惩罚函数处理约束 ,保证了算法稳定而迅速地收敛于全局最优解。本文还讨论了算法的分布式计算策略 ,并行计算是在天潮 1 0 0 0 (A)超级并行计算机上完成的 ,利用 PVM系统处理各个处理器间的通讯。算例显示了本算法的计算效率和加速比。This paper presents a parallel genetic design method with coarse grain. A new iteration scheme in conjunction with multi\|population genetic strategy, the searching technique with the narrowing down space and the quasi\|exact penalty function are employed to ensure very rapid and steady convergence. A parallel computing strategy and its implementation distributed memory, MIMD parallel computer are discussed, and a parallel program is developed on Dawning Tiangchao (1000A) parallel computer. The PVM (Parallel Virtual Machine) system is used to handle communications among processors. Numerical examples are given to demonstrate the speedup and efficiency of this method.

关 键 词：遗传算法 并行计算 粗粒度 准精确惩罚函数 多种群遗传策略 分布式计算 PVM系统 加速比 

分 类 号：O242.23[理学—计算数学]