基于alpha支配的高维目标进化算法研究

发布时间：2018-04-18 01:20

本文选题：高维目标优化 + 非支配受阻解　；参考：《计算机科学》2017年01期

【摘要】：基于Pareto支配的多目标进化算法能够很好地处理2~3维的多目标优化问题。但在处理高维多目标问题时,随着目标维数的增大,支配受阻解的数量急剧增加,导致现有的多目标算法存在选择压力不够、优化效果较差的问题。通过引入α支配提供严格的Pareto分层,在同层中挑选相对稀疏的解作为候选解,同时详细分析不同α对算法性能的影响,提出一种新的基于α偏序和拥塞距离抽样的高维目标进化算法。将该算法在DTLZ上进行性能测试,并采用世代距离(GD)、空间评价(SP)、超体积(HV)等多个指标评估算法的性能。实验结果表明,引入α支配能去除绝大部分支配受阻解(DRSs),提高算法的收敛性。与快速非支配排序算法(NSGA-II)、基于分解的多目标进化算法(MOEA/D)、基于距离更新的分解多目标进化算法(MOEA/D-DU)相比,该算法的整体解集的质量有明显提高。
[Abstract]:The multi-objective evolutionary algorithm based on Pareto can deal with the 2-dimensional multi-objective optimization problem well.However, when dealing with high dimensional multiobjective problems, with the increase of target dimension, the number of dominating obstructed solutions increases sharply, which leads to the problem that the selection pressure is not enough and the optimization effect is poor in the existing multi-objective algorithms.By introducing 伪 dominance to provide strict Pareto layering, relatively sparse solutions are selected as candidate solutions in the same layer, and the effects of different 伪 on the performance of the algorithm are analyzed in detail.A new high dimensional target evolutionary algorithm based on 伪 partial order and congestion distance sampling is proposed.The performance of the algorithm is tested on DTLZ, and the performance of the algorithm is evaluated by several indexes, such as generation distance, spatial evaluation, hypervolumic DTLZ and so on.The experimental results show that the introduction of 伪 -dominance can remove the majority of the blocked dominating solutions (DRSs) and improve the convergence of the algorithm.Compared with NSGA-IIG, multiobjective evolutionary algorithm based on decomposition and multiobjective evolutionary algorithm based on distance updating, the quality of global solution set of this algorithm is obviously improved.
【作者单位】：浙江工业大学经贸管理学院;浙江工业大学信息工程学院;浙江工业大学信息智能与决策优化研究所;
【基金】：国家自然科学基金:基于多偏好与变量分解的大规模高维目标优化方法及应用研究(61472366),基于轮廓线段簇的隐式形状模型及其优化方法研究(6139077) 浙江省自然科学基金:融合侧步爬山策略的大规模变量多目标协同进化算法研究(LY13F030010),基于双极偏好占优的高维目标进化算法研究(LZ13F020002)资助
【分类号】：TP18

【参考文献】