全局替换的自适应权重调整MOEA/D

Adaptive weight adjustment MOEA/D with global replacement

当多目标问题的帕累托前沿形状较为复杂时,基于分解的多目标进化算法MOEA/D的解的均匀性将受到很大的影响.MOEA/D利用相邻子问题的信息来优化,但早期因为种群中的个体与子问题的关联是随机分配的,仅在邻居间更新会浪费优秀解的信息,影响收敛速度.针对这些问题,本文提出一种MOEA/D的改进算法(MOEA/DGUAW).该算法使用种群全局更新的策略,来提高收敛速度;使用自适应调整权重向量的策略来获得更均匀分布的解集.将MOEA/D-GUAW算法与现有的MOEA/D,MOEA/D-AWA,RVEA和NSGA-Ⅲ算法在10个广泛应用的测试问题上进行了实验比较.实验结果表明,提出的算法在大部分问题上,反转世代距离评价指标IGD优于其他算法,收敛速度也快于其他算法.

When the shape of pareto front is complex,the uniformity of solution in multi-obective evolutionary algorithm based on decomposition(MOEA/D)will be affected.The MOEA/D uses the information of neighboring subproblems to optimize,but in the early stage,because the association between individuals in the population and subproblems is randomly assigned,updating only among neighbors will waste the information of excellent solutions and affect the convergence speed.To address these issues,an improved algorithm for MOEA/D,MOEA/D-global uniform adaptive weight(MOEA/D-GUAW),is proposed.The algorithm uses the strategy of global population update to improve the convergence speed.And the adaptive weight vector adjustment strategy are used to obtain more uniformly distributed solutions.The MOEA/D-GUAWalgorithm is compared with the existing MOEA/D,MOEA/D based on adaptive weight vector adjustment(MOEA/D-AWA),reference vector guided evolutionary algorithm(RVEA)and nondominated sorting genetic algorithm-III(NSGA-Ⅲ)in 10 widely used test problems.Experimental results show that the inverted generational distance(IGD)metric of the proposed algorithm is better than other algorithms in most problems,and the convergence speed is faster than other algorithms.