电子学报 ›› 2018, Vol. 46 ›› Issue (3): 529-536.DOI: 10.3969/j.issn.0372-2112.2018.03.003

• 学术论文 • 上一篇    下一篇

求解变量重叠型大尺度优化问题的相关性学习协同演化策略

王豫峰1,2, 董文永1, 董学士1   

  1. 1. 武汉大学计算机学院, 湖北武汉 430072;
    2. 南阳理工学院软件学院, 河南南阳 473000
  • 收稿日期:2016-09-29 修回日期:2017-01-05 出版日期:2018-03-25
    • 作者简介:
    • 王豫峰,男,1982年生于山东平度,武汉大学计算机学院博士生,研究方向为智能计算、机器学习、数据降维等.E-mail:shandian9876@163.com;董文永,男,1973年生于河南南阳,计算机软件与理论博士,现为武汉大学计算机学院教授、博士生导师.长期从事系统科学、仿真与控制、演化计算、并行计算、机器学习、数据挖掘等方面的工作.E-mail:dwy@whu.edu.cn;董学士,男,1985年出生于山东日照,武汉大学计算机学院博士生,研究方向为机器学习、智能计算等.E-mail:dxs_cs@163.com.
    • 基金资助:
    • 国家自然科学基金 (No.61170305,No.61672024); 河南省高等学校重点科研项目计划 (No.17A520046)

Cooperative Coevolution with Correlation Learning Between Variables for Large Scale Overlapping Problem

WANG Yu-feng1,2, DONG Wen-yong1, DONG Xue-shi1   

  1. 1. Computer School, Wuhan University, Wuhan, Hubei 430072, China;
    2. Software School, Nanyang Institute of Technology, Nanyang, Henan 473000, China
  • Received:2016-09-29 Revised:2017-01-05 Online:2018-03-25 Published:2018-03-25
    • Supported by:
    • National Natural Science Foundation of China (No.61170305, No.61672024); Key Scientific Research Programs of colleges and universities of Henan Province (No.17A520046)

摘要: 协同演化是解决大尺度连续优化问题的一种有效策略.但是,对于决策变量重叠型(决策变量不可分且相互依赖)的高维问题,其分组方法可能会误导算法的搜索.针对这一情况,本文提出一种全新的协同演化策略(Differential Evolution Cooperative Coevolution with Correlation Learning Between Variables,DECC-CLV),其思想是首先计算演化种群分布所包含的主特征轴,然后计算各维决策变量在主轴上的投影值并利用它们之间的正负相关性进行分组.该算法在迭代过程中,利用期望最大化算法对种群进行概率主成分分析,并根据决策变量在当前种群主轴上的投影值大小关系对其进行动态分组.通过和目前主流的演化算法在CEC2013的第三类函数上的仿真试验和分析,验证了算法的有效性和适用性.

关键词: 大尺度优化问题, 相关性决策变量, 协同演化, 大尺度优化问题分解

Abstract: Cooperative co-evolution (CC) is an effective strategy to solve large-scale continuous optimization problem. However, its grouping method may mislead the search direction when solving the large-scale overlapping problem (decision variables are non-separable and interact with each other). In order to overcome this issue, we propose a differential evolution cooperative coevolution with correlation learning between variables (DECC-CLV) to improve the performance of CC. DECC-CLV firstly detects the positive and negative correlations of variables based on the projected value of decision variables on the principal component of the population, and then groups variables into different groups. During the evolutionary process, DECC-CLV employs the expectation maximization algorithm for probabilistic principal component analysis on the population to deduce the complexity. Comparing with the state-of-the-art CCs on the large-scale overlapping benchmark functions on CEC2013, the experimental results verified the effectiveness and applicability of our proposed algorithm.

Key words: large-scale optimization problem, variables correlation, cooperative co-evolution, large-scale optimization problem decomposition

中图分类号: