基于CUDA平台的遗传算法并行实现研究.pdf

CN 4 31258/ T P 2009 3 1 A 1 ISSN 1007130X COM P U T ER EN GIN EERIN G SCIEN CE V ol3 1, NoA 1, 2009 : 1007130X ( 2009 A 1006805 * CU DA Resear ch on the Par allel Implem ent at ion of Genet ic A lgor ithm on CU DA Plat form , , TAN Caif eng, MA Anguo, XING Zuocheng ( , 410073) ( School of Compu er Science, Na ional Universi y of Defense Technology, Changsha 410073, China) : CU DA 技术方便程序员在G PU 上进行通用计算, 但并没有提供随机数产生的应用接口为 , 本文提出并 实现在CU DA 开发平台上并行产生均匀随机数算法, 测试证明算法可行在 基础上优化基本遗传算法, 并在G PU 上并 行实现其所有操作, 提高其运行速度和准确度; 分析了种群大小和遗传代数对 算法加速比及准确度的影响, 并与 M A T L A B 工具箱进行比较实验表明, 相比M AT LA B 遗传算法工具箱, 基于 CU D A 平台实现的遗传算法性能更高, 准确度更 好 Abs rac : T he CU DA technolog y pro vides co nv eniences of g enera l comput at ion for pr og rammer s, but there is no appli catio n pr og ramming interf ace of generating rando m number on CU DA . T her efor e, this paper presents and implements a met ho d fo r parallel pr oducing rando m number algo rithm o n CU DA , and the methods is prov ed f easible by t esting . On this co ndition , w e implement a parallel im plementation of G A on G PU , o ptimize the eff iciency and pr ecision o f the standard G A, analyze the influence of po pulation size and generations of evo lutio n to eff iciency and accuracy of this alg or ithm . T he ex peri ment sho w s that compared w ith GA T o olbox o f M AT L AB, the perfor mance and the precisio n o f this metho d is better : 随机数; 遗传算法; 并行计算; GP GP U ; CU DA ; G PU Key words: genetic alg or ithm ; r andom number ; parallel co mputation ; G PG PU ; CU DA ; GPU doi : 10. 3969/ j. issn. 1007130X. 2009 . A 1. 02 1 : T P30 1 : A