申威众核处理器的并行NSGA-Ⅱ算法  

Parallel NSGA-Ⅱ on Sunway many-core processor

作　　者：沈焕学 郑凯[1,2] 刘垚 王肃[1,2] 刘艳 赵瑞祥[1,2] 周谦豪 SHEN Huanxue;ZHENG Kai;LIU Yao;WANG Su;LIU Yan;ZHAO Ruixiang;ZHOU Qianhao(College of Computer and Software Engineering,East China Normal University,Shanghai 200062,China;State Key Laboratory of Mathematical Engineering and Advanced Computing,Wuxi,Jiangsu 214215,China;Faculty of Economics and Management,East China Normal University,Shanghai 200062,China)

机构地区：[1]华东师范大学计算机与软件工程学院,上海200062 [2]数学工程与先进计算国家重点实验室,江苏无锡214215 [3]华东师范大学经济与管理学部,上海200062

出　　处：《计算机工程与应用》2018年第17期35-40,共6页Computer Engineering and Applications

基　　金：数学工程与先进计算国家重点实验室开放基金(No.2016A05)

摘　　要：非支配排序遗传算法(NSGA-Ⅱ)在多目标优化领域有着广泛的应用,但在处理复杂问题时运行时间相当长。并行化是提高算法执行速度的有效途径。众核处理器的出现,为实现高度并行奠定了物质基础。基于国产超算"神威·太湖之光"的申威众核处理器平台设计了并行NSGA-Ⅱ算法(PNSGA-Ⅱ),实现了算法基于主核的一级并行和基于主/从核的二级并行。在典型测试函数集上的实验表明,在不影响解的质量前提下,PNSGA-Ⅱ算法不仅大大加快了执行速度,同时算法的收敛速度也更快。Non-dominated Sorting Genetic Algorithm-Ⅱ（NSGA-Ⅱ）is commonly used in multi-objective optimization,but featured with long time when dealing with complex problems. Algorithm parallelization can enhance the efficiency obviously, while the many-core processors provide the high-level parallelism. In this paper, a Parallel NSGA-Ⅱ algorithm（PNSGA-Ⅱ）based on Sunway many-core processor is proposed, which implements the level-1 parallelism based on master cores, and level-2 parallelism based on cooperative master/slave cores. Experimental results on typical test function sets indicate that the PNSGA-Ⅱ algorithm improves both the processing time and the convergence speed under the same accuracy.

关 键 词：非支配排序遗传算法 多目标优化 并行遗传算法 众核处理器 神威·太湖之光 

分 类 号：TP338.6[自动化与计算机技术—计算机系统结构]