ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2016, Vol. 53 ›› Issue (6): 1410-1421.doi: 10.7544/issn1000-1239.2016.20150806

• 人工智能 • 上一篇    

基于泛化反向学习的多目标约束差分进化算法

魏文红1,王甲海2,陶铭1,袁华强1   

  1. 1(东莞理工学院计算机学院 广东东莞 523808);2(中山大学计算机科学系 广州 510006) (weiwh@dgut.edu.cn)
  • 出版日期: 2016-06-01
  • 基金资助: 
    国家自然科学基金项目(61170216,61300198,61572131);广东高校科技创新项目(2013KJCX0178)

Multi-Objective Constrained Differential Evolution Using Generalized Opposition-Based Learning

Wei Wenhong1, Wang Jiahai2, Tao Ming1, Yuan Huaqiang1   

  1. 1(Computer Institute, Dongguan University of Technology, Dongguan, Guangdong 523808);2(Department of Computer Science, Sun Yat-sen University, Guangzhou 510006)
  • Online: 2016-06-01

摘要: 差分进化算法是一种简单有效的进化算法,基于泛化反向学习的机制在进化算法中经常可以引导种群的进化.针对多目标的约束优化问题,提出了一种基于泛化反向学习的多目标约束差分进化算法.该算法采用基于泛化反向学习的机制(generalized opposition-based learning, GOBL)产生变换种群,然后在种群初始化和代跳跃阶段,利用非支配排序、拥挤距离和约束处理技术从原始种群和其变换种群中选择更优的种群个体作为新的种群继续迭代进化;该算法通过采用基于泛化反向学习的机制,可以引导种群个体慢慢向最优的Pareto前沿逼近,以求得最优解集.最后采用多目标Benchmark问题对该算法进行了实验评估,实验结果表明:与NSGA-Ⅱ,MOEA/D及其他的多目标进化算法相比,提出的算法具有更好的收敛性,并且产生的解能够逼近最优的Pareto前沿.

关键词: 差分进化, 泛化反向学习, 多目标优化, 约束优化, 非支配排序

Abstract: Differential evolution is a simple and efficient evolution algorithm to deal with nonlinear and complex optimization problems. Generalized opposition-based learning (GOBL) often guides population to evolve in the evolutionary computing. However, real-world problems are inherently constrained optimization problems often with multiple conflicting objectives. Hence, for resolving multi-objective constrained optimization problems, this paper proposes a constrained differential evolution algorithm using generalized opposition-based learning. In this algorithm, firstly, the transformed population is generated using general opposition-based learning in the population initialization. Secondly, the better individuals that are selected from the initial population and the transformed population using non-dominated sorting, crowding distance sorting and constraint handling techniques compose the new initial population. Lastly, based on a jumping probability, the transformed population is calculated again after generating new populations, and the fittest individuals that are selected from the union of the current population and the transformed population compose new population using the same techniques. The solution can be evolved toward Pareto front slowly according to the generalized opposition-based learning, so that the best solutions set can be found. The proposed algorithm is tested in multi-objective benchmark problems and compared with NSGA-Ⅱ, MOEA/D and other multi-objective evolution algorithms. The experimental results show that our algorithm is able to improve convergence speed and generate solutions which approximate to the best optimal Pareto front.

Key words: differential evolution, generalized opposition-based learning (GOBL), multi-objective optimization, constrained optimization, non-dominated sorting

中图分类号: