一种基于分解和协同的高维多目标进化算法

现实中大量存在的高维多目标优化问题对以往高效的多目标进化算法提出了严峻的挑战.通过将分解策略和协同策略相结合提出一种高维多目标进化算法MaOEA/DCE.该算法利用混合水平正交实验方法在聚合系数空间产生一组均匀分布的权重向量以改善初始种群的分布性;其次,算法将差分进化算子和自适应SBX算子进行协同进化,以产生高质量的子代个体,并改善算法的收敛性.该算法与另外5种高性能的多目标进化算法在基准测试函数集DTLZ{1,2,4,5}上进行对比实验,利用改进的反转世代距离指标IGD+评估各算法的性能.实验结果表明,Ma OEA/DCE算法与其他对比算法相比,在总体上具有较为显著的收敛性和分布性优势.

Many-objective Evolutionary Algorithm Based on Decomposition and Coevolution

In reality, many-objective optimization problems (MaOPs) are ubiquitous, and their huge objective space makes representative multi-objective evolutionary algorithms face severe challenges. A many-objective evolutionary algorithm based on decomposition and coevolution (MaOEA/DCE) is proposed in the paper. The method of mix-level orthogonal experimental design is used to generate an even-distributed set of weight vectors in the space of weight coefficient in order to improve the population diversity. Second, the differential evolution operator is combined with the adaptive SBX operator to generate high-quality offspring, so as to improve the convergence of MaOEA/DCE. The MaOEA/DCE is compared with other five representative multi-objective evolutionary algorithms on the benchmark test set DTLZ{1,2,4,5} based on the performance of IGD+, and the experimental results show that the presented algorithm has more significant performance advantages of convergence, diversity and robustness over the peering algorithms.