摘要
现代飞机设计对气动外形优化效率的要求不断提高,传统气动力获取方法(如风洞试验或计算流体力学(CFD)数值仿真方法)成本高、效率低,探索高效的气动力获取方法对减少风洞试验或数值仿真成本、提高飞机迭代设计效率具有重要意义。本文提出一种基于集成学习的飞机气动力快速预测方法,将线性回归模型、多层感知机模型、梯度提升模型堆叠,对不同机翼展长、根弦比、尖弦长的飞翼布局无人机在不同迎角下的气动力系数进行预测。结果表明,建立的集成学习模型能够快速准确预测飞机气动力系数,测试集升阻力系数均方误差分别为0.208×10^(-4)和0.424×10^(-5),平均绝对误差分别为0.27×10^(-2)和0.1379×10^(-2),拟合度分别为0.9994976和0.9691,预测时间为0.8s,仅为面元法计算时间的1/4500,有效地提高了飞机气动外形设计效率。
The demand for aerodynamic shape optimization efficiency in modern aircraft design is constantly increasing.Traditional aerodynamic force acquisition methods such as the wind tunnel experiment or the CFD numerical simulation have high costs and low efficiency.Exploring efficient aerodynamic force acquisition methods is of great significance in reducing wind tunnel testing or numerical simulation costs and improving aircraft iterative design efficiency.A fast prediction method for aircraft aerodynamics based on ensemble learning is proposed in this article.The linear regression model,multi-layer perceptron model,and gradient boosting model are stacked to predict the aerodynamic force coefficients of the flying wing layout drones with different wing span lengths,root chord ratios,and tip chord lengths at different angles of attack.The results show that the established ensemble learning model can predict the aerodynamic coefficients of aircraft quickly and accurately.The mean square errors of the lift and drag coefficients in the test sets are 0.208×10^(-4) and 0.424×10^(-5),respectively,with the mean absolute errors of 0.27×10^(-2) and 0.1379×10^(-2),the fitting degrees of 0.9994976 and 0.9691,and a prediction time of 0.8s,which is only 1/4500 of the calculation time of the panel method,which improves the efficiency of aircraft aerodynamic shape design effectively.
作者
刘哲
郭承鹏
李鸿岩
崔榕峰
Liu Zhe;Guo Chengpeng;Li Hongyan;Cui Rongfeng(Aviation Key Laboratory of Science and Technology on Aerodynamics of High Speed and High Reynolds Number,AVIC Aerodynamics Research Institute,Shenyang 110034,China)
出处
《航空科学技术》
2024年第11期13-18,共6页
Aeronautical Science & Technology
基金
航空科学基金(2022Z006026004,2023M071027001)。
关键词
气动力
集成学习
快速预测
梯度提升模型
堆叠法
aerodynamics
ensemble learning
quick prediction
Gradient Boosting model
stacking method
