J4 ›› 2014, Vol. 41 ›› Issue (1): 98-104+188.doi: 10.3969/j.issn.1001-2400.2014.01.018

• 研究论文 • 上一篇    下一篇

求解约束优化问题的偏好多目标进化算法

董宁1;王宇平2   

  1. (1. 西安电子科技大学 数学与统计学院,陕西 西安  710071;
    2. 西安电子科技大学 计算机学院,陕西 西安  710071)
  • 收稿日期:2012-10-16 出版日期:2014-02-20 发布日期:2014-04-02
  • 通讯作者: 董宁
  • 作者简介:董宁(1980-),女,西安电子科技大学博士研究生,E-mail: dongning@snnu.edu.cn.
  • 基金资助:

    国家自然科学基金资助项目(61272119)

Multi-objective evolutionary algorithm based on preference for constrained optimization problems

DONG Ning1;WANG Yuping2   

  1. (1. School of Mathematics and Statistics, Xidian Univ., Xi'an  710071, China;
    2. School of Computer Science and Technology, Xidian Univ., Xi'an  710071,China)
  • Received:2012-10-16 Online:2014-02-20 Published:2014-04-02
  • Contact: DONG Ning

摘要:

将约束优化问题转化为双目标优化问题,用进化算法求解转化的双目标问题.设计了新的混合交叉算子以提高算法在进化过程中的搜索能力,加快算法收敛;借鉴多目标优化加权度量法中成绩标量函数的特点,提出新的偏好适应度函数,进行个体比较和选择.新适应度以个体到参考点的加权距离衡量个体优劣,参考点和权向量体现选择的偏好.在进化过程中,自适应地选择参考点和权向量平衡进化的不同阶段对各个目标的偏好程度,增加种群多样性,避免算法早熟收敛.

关键词: 约束优化, 多目标优化, 进化算法, 偏好, 成绩标量函数

Abstract:

Constrained optimization problems (COPs) are converted into the bi-objective optimization problem and solved with a new preference based multi-objective evolutionary algorithm. A new hybrid crossover operator is proposed to improve the search ability in the evolutionary process, and also a novel fitness function with preference based on the achievement scalarizing function (ASF) which is used in the method of weighted metrics in multi-objective optimization is presented. The new fitness measures the merits of individuals by the weighting distance from individuals to the reference point, where the reference point and the weighting vector afford the preference for selection. In different evolutionary stages, the reference point and weighting vector are chosen adaptively according to the individuals in population to make a tradeoff between the preferences to the two objectives. Numerical experiments for several standard test functions with different characteristics illustrate that the new proposed algorithm is effective and efficient.

Key words: constrained optimization, multi-objective optimization, evolutionary algorithm, preference, achievement scalarizing function

中图分类号: 

  • TP301. 6