基于XGBoost的多源气动数据融合建模

Multi-source aerodynamic data fusion modeling with XGBoost

飞行试验和计算流体力学(CFD)是气动数据获取的两种常用手段,其中,通过飞行试验获取的气动数据精度高,但成本高昂,而通过CFD获取气动数据成本低,数据量丰富。因此,为了以尽可能低的成本获取精度更高的气动数据,本文利用数据融合方法,提出了一种可以融合大量的由CFD获取的气动数据与少量由飞行试验获取的气动数据的嫁接XGBoost(eXtreme Gradient Boosting)集成模型框架。首先,利用充足的CFD气动数据建模,作为融合框架中的低精度模型;然后,将少量飞行试验参数输入低精度模型得到相应的气动输出;最后,结合飞行试验气动数据的其他特征进行二次建模,实现对真实气动参数的预测。为了证明所提算法的有效性,本文设置了相关的对比实验,结果表明:1)多源气动数据融合建模较单源气动数据建模具有更高的预测精度;2)集成学习模型比传统的机器学习模型具有更强的泛化性能,其中XGBoost模型泛化能力最强,能够实现对气动参数的准确预测。

Flight tests and CFD(Computational Fluid Dynamics)are frequently used to obtain aerodynamic data.The aerodynamic data obtained by flight tests are usually highly accurate but expensive,while those obtained by CFD are relatively less accurate,but cheap and abundant in data.Therefore,to obtain aerodynamic parameters with higher accuracy at the lowest possible cost,this paper proposes a grafted XGBoost(eXtreme Gradient Boosting)integrated model framework,combining aerodynamic data obtained by CFD and flight tests for fusion modeling.First,a large amount of CFD aerodynamic data are used to establish a low-precision model in the fusion modeling framework regarding the trend of aerodynamic data.Then,a small number of design variables of flight tests are put into the low-precision model to obtain the corresponding aerodynamic output.Finally,the secondary modeling is carried out by integrating the information from other flight test aerodynamic data characteristics to predict the exact aerodynamic output.In order to verify the effectiveness of the proposed algorithm,relevant comparative experiments are conducted.The results show that:1)Multi-source aerodynamic data fusion modeling has higher prediction accuracy than its single-source counterparts;2)The integrated learning model has a better generalization performance than traditional machine learning models,among which the XGBoost model has the best generalization ability and can realize the accurate prediction of aerodynamic parameters.