论文部分内容阅读
多目标优化问题是进化算法领域的研究热点与难点.基于分解的多目标进化算法(MOEA/D)在求解多目标优化问题时有着较强的搜索能力、高效的适应度评价、良好的收敛性等优点.然而,不同的子问题使用相同大小的邻域统一优化,减缓算法搜索全局最优解的速率.为解决以上问题,提出一种动态邻域设置策略,针对不同的子问题设置不同的邻域.首先,分析子问题差异处理的原因;其次,根据子问题与边界的距离,提出边界子问题与靠边界子问题的邻域减小,其他子问题邻域增大策略并将以上策略应用在MOEA/D中,提出一种动态邻域的分解多目标进化算法,进一步分析改进算法中参数的敏感性.将该算法在经典测试函数ZDT系列,WFG系列上进行仿真实验,并采用反向世代距离(IGD)和超体积(HV)指标对算法性能对比分析.结果表明,与MOEA/D对比,改进算法的收敛性明显提高,求出的解集相比MOEA/D,NSGA-II,MOEA/D-DU同类典型的算法求出解集的质量更高,算法在求解前端为凸面的情况效果甚好.
Multi-objective optimization is a hot and difficult topic in the field of evolutionary algorithms.The decomposition-based multi-objective evolutionary algorithm (MOEA / D) has strong search ability, efficient fitness evaluation and good convergence in solving multi-objective optimization problems Etc. However, different sub-problems use uniform optimization of the same size neighborhood to reduce the rate at which the algorithm searches for the global optimal solution.In order to solve the above problem, a dynamic neighborhood setting strategy is proposed to set different strategies for different sub-problems First of all, the paper analyzes the reasons of the sub-problem difference processing. Secondly, according to the distance between the sub-problem and the boundary, the neighborhood of the boundary sub-problem and the boundary-dependent subproblem are reduced, and the neighborhoods of other sub-problems are increased. In MOEA / D, a dynamic neighborhood decomposition multi-objective evolutionary algorithm is proposed to further analyze the sensitivity of the improved algorithm. The algorithm is simulated on the classic test function ZDT series and WFG series, The results show that compared with MOEA / D, the convergence of the improved algorithm is obviously improved, and the solution set phase Higher MOEA / D, NSGA-II, MOEA / D-DU exemplary algorithm similar set of solutions is obtained by mass, the algorithm for solving a convex front end of the case where the effect is very good.