超参数自适应的MOEA/D-DE算法在翼型气动隐身优化中的应用

Application of hyper-parameters adaptive MOEA / D-DE algorithm in aerodynamic stealth optimization of airfoil

摘要:

MOEA/D-DE算法易于实现，被广泛应用于处理多目标优化问题，可是其超参数CR和F对算法性能影响较大。本文基于MOEA/D-DE算法框架利用Sobol全局灵敏性分析方法对差分进化算子中的交叉控制参数CR进行改进，然后使用莱维飞行策略控制比例因子F，使算法中的超参数拥有自适应能力，最后得到了超参数自适应的MOEA/D-DE算法MOEA/D-DEAH。对MOEA/D-DEAH、不同超参数设置的MOEA/D-DE算法和NSGAII算法进行了函数测试和翼型气动隐身优化算例对比，函数测试结果显示新算法性能良好，通过对超参数进行自适应可以使算法获得了较强的鲁棒性。翼型气动隐身优化算例结果显示MOEA/D-DEAH算法效果较其他算法优化结果更好。

Abstract:

MOEA/D-DE algorithm is easy to implement and widely used to deal with multi-objective optimization problems, while its hyper-parameters CR and F have a great impact on the performance of the algorithm. In this paper, based on the MOEA/D-DE algorithm framework, the Sobol global sensitivity analysis method is used to improve the cross control parameter CR in the differential evolution operator, and the Levy flight strategy is used to control the scale factor F. Then the hyper-parameters in the MOEA/D-DE algorithm obtain adaptive ability. And MOEA/D-DE algorithm with adaptive hyper-parameters (MOEA/D-DEAH) is proposed. The MOEA/D-DEAH MOEA/D-DE with different hyper-parameters settings and NSGAII are tested and compared with function tests and aerodynamic stealth optimization of airfoil. The function test results show that the new algorithm has good performance, and the algorithm can obtain strong robustness by adapting the super parameters. The optimization results of airfoil aerodynamic stealth show that MOEA/D-DEAH algorithm has better optimization results than other algorithms.