基于贝叶斯超参数优化的Gradient Boosting方法的导弹气动特性预测

Prediction of Missile Aerodynamic Data Based on Gradient Boosting under Bayesian Hyperparametric Optimization

在导弹设计与研发的初期阶段,需要寻求高效且低成本的导弹气动力特性的分析方法。然而,气动性能分析过程中往往存在试验成本高、周期长、局限性大等问题。因此,本文采用基于提升(Boosting)的机器学习集成算法进行导弹气动特性预测,通过输入导弹的气动外形参数、马赫数和迎角,对于导弹气动力系数实现快速预测。结果表明,Boosting能够对导弹气动力系数进行精准高效预测。为进一步提升预测精度,与传统的机器学习参数调整方法相比,采用贝叶斯优化方法对梯度提升(Gradient Boosting)算法超参数进行优化,调优后的Gradient Boosting方法预测的导弹气动力系数与实际值吻合度得到提升,并将贝叶斯优化的Gradient Boosting方法与XGBoost、LightGBM、Adaboost方法进行了对比,贝叶斯优化的Gradient Boosting方法预测精度优于其他Boosting方法,证明了优化方法的可行性与有效性。

In the initial stage of missile design,the aerodynamic characteristics of missiles need to be quickly and preliminarily evaluated.When analyzing the aerodynamic performance of missiles,the traditional engineering method has the problem of high experiment costs and the long experiment period.Meanwhile,the CFD method is also difficult to calculate due to its complex calculation process,and the calculation cost is high.Therefore,this paper applys the method of missile aerodynamic data prediction based on several Boosting methods in machine learning.By inputting aerodynamic shape,mach number,and angle of attack data of the missile,the lift coefficient and drag coefficient are quickly predicted.The result shows that Boosting method can predict the aerodynamic coefficients of missiles accurately.In order to further improve the prediction accuracy,compared with other traditional hyperatameter selection methods,Bayesian Hyperparameter Optimization method as an automatic hyperatameter selection method is used to optimize the parameters of the Gradient Boosting algorithm,and it turns out that the predicted value is much closer to the actual value.Finally,the Gradient Boosting method under Bayesian Optimization is compared with XGBoost,LightGBM and Adaboost methods,and the Gradient Boosting method under Bayesian Optimization is more accurate than other Boosting methods,which proves the feasibility and effectiveness.