航空学报 > 2021, Vol. 42 Issue (4): 524093-524093   doi: 10.7527/S1000-6893.2020.24093

基于自动核构造高斯过程的导弹气动性能预测

胡伟杰1,2, 黄增辉3, 刘学军1,2, 吕宏强3   

  1. 1. 南京航空航天大学 计算机科学与技术学院/人工智能学院, 模式分析与机器智能工业和信息化部重点实验室, 南京 211106;
    2. 软件新技术与产业化协同创新中心, 南京 210023;
    3. 南京航空航天大学 航空学院, 南京 210016
  • 收稿日期:2020-04-15 修回日期:2020-07-07 发布日期:2020-07-17
  • 通讯作者: 刘学军 E-mail:xuejun.liu@nuaa.edu.cn
  • 基金资助:
    航空科学基金(2018ZA52002,2019ZA052011);空气动力学国家重点实验室基金(SKLA20180102);气动噪声控制重点实验室基金(ANCL20190103)

Missile aerodynamic performance prediction of Gaussian process through automatic kernel construction

HU Weijie1,2, HUANG Zenghui3, LIU Xuejun1,2, LYU Hongqiang3   

  1. 1. MⅡT Key Laboratory of Pattern Analysis and Machine Intelligence, College of Computer Science and Technology/College of Artificial Intelligence, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China;
    2. Collaborative Innovation Center of Novel Software Technology and Industrialization, Nanjing 210023, China;
    3. College of Aerospace Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China
  • Received:2020-04-15 Revised:2020-07-07 Published:2020-07-17
  • Supported by:
    Aeronautical Science Foundation of China (2018ZA52002,2019ZA052011); Foundation of State Key Laboratory of Aerodynamics (SKLA20180102); Key Laboratory of Aerodynamic Noise Control (ANCL20190103)

摘要: 在导弹的初期设计阶段,通常需要对导弹的气动性能进行快速粗略评估。针对传统工程估算软件计算精度低和CFD方法计算代价大的缺陷,提出一种基于高斯过程回归(GPR)代理模型快速预测典型导弹气动性能的方案。以导弹外形参数和攻角作为模型输入,升力系数、阻力系数和力矩系数作为模型输出,对GPR模型的气动性能预测结果进行分析。首先,与其他常用代理模型的预测精度对比,GPR模型对3种系数的预测误差分别仅为0.24%、0.36%和0.36%,高于其他代理模型的预测精度。其次,考虑GPR模型核函数选择严重依赖人工先验知识的问题,采用了一种自动核构造算法,无需先验知识即可从数据中自动学习核函数。将该算法嵌入GPR框架中,与传统GPR模型比较,实验结果表明:基于该算法的GPR模型对3种系数的预测误差分别降低到0.10%、0.22%和0.17%。最后,给出导弹气动性能快速预测的应用实例,结果表明所提出的GPR模型的导弹气动性能预测方案是有效的,能够满足设计初期快速且精确的气动性能预测需求。

关键词: 导弹气动性能, 预测, 代理模型, 高斯过程回归, 自动核构造, 核函数

Abstract: The initial stage of missile design usually requires quick and rough evaluation of missile aerodynamic performance. To improve the low calculation accuracy of traditional engineering estimation software and reduce the high calculation cost of CFD method, we propose a scheme based on the Gaussian Process Regression (GPR) surrogate model to quickly and accurately predict the aerodynamic performance of typical missiles. The prediction results of the GPR model are analyzed, taking the missile shape parameters and angle of attack as the input and the lift coefficient, drag coefficient and moment coefficient as the output. First of all, compared with the prediction accuracy of other commonly used surrogate models, the prediction errors of the GPR model for the three coefficients are only 0.24%, 0.36% and 0.36%, respectively, lower than those of other surrogate models. Secondly, considering the problem that the kernel function selection of the GPR model depends heavily on artificial prior knowledge, we embed an automatic kernel construction algorithm which can automatically learn the kernel function from data without prior knowledge into the GPR framework. Compared with the traditional GPR model, the prediction errors of the GPR model incorporating the algorithm for the three coefficients are reduced to 0.10%, 0.22% and 0.17%, respectively. Finally, an application example of rapid prediction of missile aerodynamic performance is conducted, and the results show that the prediction scheme of the missile aerodynamic performance based on the GPR model is effective and can meet the requirements of fast and accurate aerodynamic performance prediction in the initial stage of missile design.

Key words: missile aerodynamic performance, prediction, surrogate models, Gaussian process regression, automatic kernel construction, kernel function

中图分类号: