应用档案精英学习和反向学习的多目标进化算法

Multi-Objective Evolutionary Algorithm Based on Archive-Elite Learning and Opposition-Based Learning

现实中的多目标优化问题日益复杂,对多目标优化算法提出了新的挑战.受混合多目标优化算法的启发,该文提出了一种应用档案精英学习和反向学习的多目标进化算法(Multi-objective Evolutionary Algorithm Based on Archive-Elite Learning and Opposition-based Learning,AOL-MOEA)以解决困难的多目标优化问题.AOLMOEA算法利用档案精英学习算子增强算法全局搜索能力,促进算法较快收敛;运用动态一般反向学习机制代替变异算子以增加种群逃逸局部极值的机会;使用3-点最短路径方法维持解群的多样性.AOL-MOEA算法与另外5种代表性多目标优化算法在12个基准多目标测试函数上进行性能比较,实验结果表明:AOL-MOEA算法在收敛性、多样性和稳定性等方面均优于或部分优于其他的对比算法.

It is a huge challenge for multi-objective optimization algorithms due to the increasing complexity of the multi-objective optimization problems（MOPs for short）in the real world.Inspired by the idea of hybrid components of multi-objective optimization algorithms,a new multi-objective evolutionary algorithm based on archive-elite learning and opposition-based learning（AOL-MOEA for short）was proposed to tackle some complicated MOPs in the paper.The proposed AOL-MOEA used the strategy of archive-elite learning to enhance the ability of global search so as to promote the convergence of the algorithm.Secondly,a dynamic generalized opposition-based learning approach was utilized to replace the traditional mutation operator to increase the probability of escaping from local optima for the optimizer.Thirdly,a novel diversity preserved mechanism of three-point shortest path was proposed to conquer some intrinsic defects of the popular diversity preserved strategies,and maintenance the diversity of the population effectively.The AOL-MOEA was compared with other five typical multi-objective optimization algorithms on a benchmark test set including 12 multi-objective optimization functions.Experimental results demonstrate that the proposed algorithm outperforms or partially outperforms the other peer algorithms on convergence,diversity and stability.