摘要
设计变量“维数灾难”问题是限制基于代理模型优化算法应用到高维气动设计的关键技术难题。为了解决维度灾难问题导致的代理模型精度下降和优化效果变差的问题,改进一种基于物理知识驱动的敏感性分区代理模型优化方法,研究不同分区设计变量对目标函数的敏感性;在序贯分区优化的基础上,以敏感性作为分区代理模型优化的顺序,开展进一步的优化设计研究。结果表明:该方法将高维设计空间分解为一系列低维子空间,可以显著提高代理模型的预测精度,实现样本的高效配置,从而实现高效的全局搜索,相较于传统代理优化方法,其建立代理模型所花费时间大幅减少,提高了代理优化方法处理高维气动设计问题的能力。
The problem of"dimensionality disaster"of design variable is a key technical problem that restricts the application of current agent-based optimization algorithms.In order to solve the problem of declining accuracy and poor optimization effect of surrogate model caused by dimensionality disaster problem,a sensitivity analysis partition surrogate model optimization method based on physical knowledge driven is improved.The sensitivity of different partition design variables to the objective function is studied.On the basis of the sequential partition optimization,the sensitivity is taken as the order of the partition surrogate model optimization.The results show that the method can solute the high-dimensional design space into a series of low-dimensional subspaces,improve the prediction accuracy of surrogate model,realize the efficiency configuration,so as to realize the global search.In comparison with traditional surrogate optimization method,the time spent on establishing the surrogate model is much lower,and the ability of surrogate optimization method to solve the high-dimensional aerodynamics design problem.
作者
邵梦莹
夏露
张伟
赵轲
SHAO Mengying;XIA Lu;ZHANG Wei;ZHAO Ke(School of Aeronautics,Northwestern Polytechnical University,Xi’an 710072,China;National Key Laboratory of Aircraft Configuration Design,Xi’an 710072,China)
出处
《航空工程进展》
2025年第4期48-62,92,共16页
Advances in Aeronautical Science and Engineering
基金
飞行器基础布局全国重点实验室稳定支持项目(614220121020128)
飞行器基础布局全国重点实验室基金(JBGS-2024-02,2023-JCJQ-LB-070)。
关键词
分区优化
高效全局优化方法
维数灾难
敏感性分析
代理模型精度
partition optimization
efficient global optimization method
dimensionality disaster
sensitivity analysis
accuracy of surrogate model
