火箭弹气动学科近似方法分析  

Approximation Methods Analysis of Rocket Aerodynamics

作　　者：赵良玉[1] 杨树兴[1] 佘浩平[1] 

机构地区：[1]北京理工大学宇航科学技术学院,北京100081

出　　处：《弹箭与制导学报》2008年第3期175-180,共6页Journal of Projectiles,Rockets,Missiles and Guidance

基　　金：国防基础科研基金资助

摘　　要：为了确定最适合于火箭弹气动学科的近似方法,分别采用二阶响应面方法、Kriging方法、径向基神经网络方法构建了火箭弹气动学科代理模型,并对三种代理模型的近似精度做了比较。针对现有近似方法评估策略的不足,提出了一种基于相对误差综合均值的近似效果综合评估策略。结果表明该评估策略是有效的,二阶响应面近似方法不可取,径向基神经网络近似方法部分点数据的近似精度低于Kriging方法,但其综合性能最好。The applicability of different aerodynamic sub-discipline approximation method of rocket aerodynamic sub-discipline surrogate model establishment was studied. The rocket aerodynamic Sub-discipline surrogate models were established by means of square response surface method, Kriging method and RBF neural network respectively, and the accuracy of each surrogate model was compared. An integrative evaluating strategy based on the comprehensive mean value of the relative error was introduced to consummate the existing evaluating strategies. Subsequently, the precision of three approximation method was evaluated. The result shows that this evaluating method is effective, 2nd order response surface method is not advisable, and through several results of RBF NN are worse than Kriging method in some points, its comprehensive performance is the best.

关 键 词：近似方法 火箭弹 RBF神经网络 

分 类 号：TJ415[兵器科学与技术—火炮、自动武器与弹药工程]