基于alpha支配的高维目标进化算法研究

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。

Research of Many-objective Evolutionary Algorithm Based on Alpha Dominance

The classic multiobjective evolutionary algorithms based on the Pareto dominance solve the problems with 2 to 3 objectives effectively.However,when dealing with many-objective problems,as the number of dominance resistant solutions is rapidly increasing owing to the increase of the objectives,the existed multiobjective algorithms lack of the selection pressure towards the Pareto front,and the optimization effect becomes bad.In this paper,we analyzed the influence of different alpha values,then provided strict Pareto layer,and selected the relatively sparse solution as candidate solutions in the same layer.At last,we proposed a new many-objective evolutionary algorithm based on alpha partial order and congestion distance sampling.The performance of the algorithms was evaluated by generation distances(GD),spacing(SP),highpervolume(HV) on the DTLZ problems.The experimental results show that the convergence of the algorithm improves greatly through eliminating the DRSs.Compared with the NSGA-II,MOEA/D and MOEA/D-DU,the overall quality of the solutions by the improved algorithms increases greatly.