基于机器学习的巡飞弹气动优化与制导一体化设计
Machine learning-based integrated design to aerodynamic optimization and guidance for cruise missile
吴明雨 1何贤军 1郑纯 1陈志华2
作者信息
- 1. 南京理工大学 能源与动力工程学院,南京 210094
- 2. 南京理工大学 瞬态物理重点实验室,南京 210094
- 折叠
摘要
针对巡飞弹末制导过程,提出了一种基于机器学习的气动优化与制导一体化设计方法.该方法首先以巡飞弹机翼为自由变形对象建立了基于全连接神经网络的气动参数代理模型,结合该模型提出了物理参数优化神经网络进行气动寻优,以此构建了在线气动优化模型;其次,耦合该优化模型和巡飞弹飞行动力学方程组,搭建了强化学习的制导环境,利用深度Q网络强化学习算法设计了智能制导律,实现了气动优化与制导一体化设计.通过打击地面静止目标的仿真实验,与无气动优化的制导律相比,表明:在线气动优化可使得巡飞弹以最优升阻比飞行,缩短了制导时间,节约燃料;一体化模型仅利用弹目视线角速率信息,可产生较连续的攻角制导指令,提高了制导精度.
Abstract
A machine learning-based aerodynamic optimization and guidance integration design method is proposed for the terminal guidance of the cruise missile.Taking the cruise missile's wing as the optimization object,the method firstly establishes a fully connected neural network-based aerodynamic parameter surrogate model,and combines the model with a physical parameter optimization neural network for aerodynamic optimization,thus constructing an online aerodynamic optimization model.Secondly,the guidance environment of reinforcement learning is constructed by coupling the optimization model with the flight dynamics equation set of the cruise missile,and an intelligent guidance law is designed by using the Deep Q-Network reinforcement learning algorithm,thus realizing the integrated design of aerodynamic optimization and guidance.Through the simulation experiments of hitting the ground stationary target,compared with the guidance law without aerodynamic optimization,it shows that online aerodynamic optimization can make the cruise missile fly with the optimal lift-to-drag ratio,which shortens the guidance time and saves the fuel;the integrated model can produce more continuous angle-of-attack guidance commands by using only the line-of sight angle rate,which improves the guidance accuracy.
关键词
气动优化/制导律/代理模型/一体化设计/机器学习/深度Q网络Key words
aerodynamic optimization/guidance law/surrogate model/integrated design/machine learning/Deep Q-Network引用本文复制引用
出版年
2024
NETL
NSTL
万方数据