基于PINN模型的导弹气动特性快速预测技术  

Rapid prediction technology of missile aerodynamic characteristics based on PINN model

作　　者：蔺佳哲 周岭[1] 武频[2] 袁雯琰 周铸[1] LIN Jiazhe;ZHOU Ling;WU Pin;YUAN Wenyan;ZHOU Zhu(Computational Aerodynamic Research Institute,China Aerodynamics Research and Development Center,Mianyang 621000,China;School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China)

机构地区：[1]中国空气动力研究与发展中心计算空气动力研究所,绵阳621000 [2]上海大学计算机工程与科学学院,上海200444

出　　处：《北京航空航天大学学报》2023年第10期2669-2678,共10页Journal of Beijing University of Aeronautics and Astronautics

摘　　要：随着内嵌物理机理神经网络(PINN)模型的兴起,PINN模型开始应用于许多学科领域。为了实现导弹气动特性的快速预测,借助工程算法,构建了导弹气动数据集,以此训练导弹气动特性预测模型,包含基于多任务学习的神经网络(MTLNN)模型及在MTLNN模型基础上内嵌物理知识的PINN模型。数值模拟通过选取测试集,对比了MTLNN模型和PINN模型的预测效果,结果表明:PINN模型的预测精度较高,且基本控制在1%以内。探究PINN模型的泛化能力,测试集选取导弹气动数据集包络范围之外的数据,PINN模型预测精度仍然高于MTLNN模型。由于PINN模型引入了气动特性参数之间的物理机理,模型对训练样本数量的依赖程度降低,可以进一步节约数据获取成本,为导弹优化设计提供有力工具。With the rise of the physical-informed neural network(PINN)model,the PINN model has been applied to many subjects.With the aid of the missile engineering algorithm,the missile aerodynamic data set is created in order to train the multi-task learning neural network(MTLNN)model and the physical-informed-PINN model,two models that can quickly predict missile aerodynamic characteristics.By selecting test sets,the numerical simulation compares the prediction results of the MTLNN model with the PINN model,and the result shows that the prediction accuracy of the PINN model is higher,and the prediction relative error is less than 1%.Finally,the generalization ability of PINN model is explored.The test set selects data outside the envelope range of the missile aerodynamic data set.In this case,the prediction accuracy of the PINN model is higher than that of the MTLNN model.The PINN model has a physical mechanism connecting the parameters that control aerodynamic properties,which makes the model less reliant on the volume of training samples.This can further reduce data collection costs and give a strong tool for missile optimization design.

关 键 词：内嵌物理机理神经网络 导弹 气动特性 快速预测 数据驱动 

分 类 号：V211.24[航空宇航科学与技术—航空宇航推进理论与工程] TJ760.11[兵器科学与技术—武器系统与运用工程]