J4 ›› 2010, Vol. 48 ›› Issue (05): 787-792.

• 计算机科学 • 上一篇    下一篇

基于CMP的多种并行蚁群算法及比较

何丽莉, 王克淼, 白洪涛, 胡成全   

  1. 吉林大学 计算机科学与技术学院, 长春 130012|吉林大学 符号计算与知识工程教育部重点实验室, 长春 130012
  • 收稿日期:2009-08-23 出版日期:2010-09-26 发布日期:2010-09-21
  • 通讯作者: 何丽莉 E-mail:helili@jlu.edu.cn

Multi Parallel Ant Colony Optimization AlgorithmsBased on CMP and Their Comparison

HE Li li, WANG Kemiao, BAI Hongtao, HU Chengquan   

  1. College of Computer Science and Technology, Jilin University, Changchun 130012, China|Key Laboratory of |Symbolic Computation and Knowledge Engineering of Ministry of Education, Jilin University, Changchun 130012, China
  • Received:2009-08-23 Online:2010-09-26 Published:2010-09-21
  • Contact: HE Li li E-mail:helili@jlu.edu.cn

摘要:

基于片上多核处理器(Chip Multiprocessor, CMP)的多种并行蚁群算法, 包括并行最大最小蚂蚁系统、 并行蚁群系统及两者的混合等5个并行算法, 提出一种在CMP的每个处理器核心上模拟一个子蚁群, 整体蚁群共享同一信息素矩阵, 实现信息素隐式交流的方法. 用多线程实时优先级实现该算法, 并用若干旅行商问题实例进行了测试, 分析了不同并行策略的影响. 测试结果表明, 基于CMP的并行蚁群具有相对于核心数目的线性加速比, 异种蚁群混合策略在解的稳定性上更具优势.

关键词: 蚁群优化; 共享信息素矩阵; 并行计算; 片上多核处理器

Abstract:

Multi parallel ant colony optimization algorithms based on chip multiprocessor (CMP) proposed in this paper include parallel MAXMIN ant system, parallel ant colony system and diffrent mixture strategies of both the systems. Each core of the CMP simulates a subant colony and the whole ant system shares one pheromone matrix. The implementation was obtained by the multi
threads with real time priority. The influence of the strategies was examined by several traveling salesman problem benchmarks from TSPLIB. Experiments show that parallel ACOs based on CMP have nearly linear speedup by the number of cores and the mixture of MAXMIN ant system and ant colony system has the advantage in the stability of solutions.

Key words: ant colony optimization, sharing one pheromone matrix, parallel computing, chip multiprocessor

中图分类号: 

  • TP18