在超目标优化问题中，目标之间的冲突会导致没有一个解可以同时优化所有目标，求解时存在大量非支配解。选择合适的解排序算法评估解的质量，对算法性能起着关键作用。而不同的解排序算法，在处理不同的超目标问题时有着各自的优劣。因此，本文提出一个基于投票机制和动态分配价值点的集成框架(ensemble many-objective evolutionary algorithm based on voting and dynamic value point，VDVP-EMEA)，将不同解排序算法聚合在一起协同工作。首先，根据每种解排序算法的有效投票率，动态分配每个专家拥有的价值点，有效投票越多的解排序算法，相应赋予更多的价值点，反之则对价值点进行惩罚。然后使用末位淘汰制，废弃能力最差的专家的投票。其次，在环境选择过程中，使用精英选择策略，通过投票结果和价值点来定义个体适应度，适应度越大的个体越优先被选择。最后，为了测试 VDVP-EMEA 算法的性能，进行大量试验，将 VDVP-EMEA 与 4 种常用的单一解排序算法NSGA-III、SPEA2、BiGE、GrEA和一种先进的集成算法VMEF进行了比较。实验结果表明，VDVP-EMEA的收敛性和多样性明显优于这些算法。
An Ensemble Many-objective Optimization Algorithm Based on Voting and Dynamic Value Point
In many-objective optimization problems,the conflicts among the objectives lead to the situation where solutions cannot optimize all objectives simultaneously,and a large number of non-dominated solutions exist in the process.An appropriate solution-sorting method plays a crucial role in evaluating the quality of solutions and the performance of the algorithm.Generally,different solution-sorting methods have their own pros and cons when handling different many-objective problems.Therefore,an ensemble many-objective evolutionary algorithm based on voting and dynamic value point(VDVP-EMEA)is proposed,which can integrate different solution-sorting methods and cooperate together.First,the value points of each expert are dynamically allocated by the voting success rate of each solution-sorting method,and the solution-sorting method with more voting success rate will be correspondingly assigned more value points.Otherwise,the value points will be punished.The last elimination system is used to cancel the votes of the most inefficient experts.The elite selection strategy is used to define fitness of individuals by voting results and value points,and the individuals with greater fitness value are preferentially selected in the process of environment selection.Finally,a large number of experiments are conducted to test the performance of VDVP-EMEA,and VDVP-EMEA is compared with five advanced many-objective evolutionary algorithms NSGA-III,SPEA2,BiGE,GrEA,and VMEF.Experimental results indicate that the overall performance of VDVP-EMEA is distinctly better than these algorithms.
万方数据
