基于神经网络的飞行器控制方法及仿真研究
|
摘要
大气层内飞行器控制问题是典型的高不确定性、严重非线性、强耦合、快时变的多变量系统控制问题,传统"前馈+自适应PD"控制律设计过程中面临配平舵偏角、气动偏导数等气动参数超曲面拟合问题,主要采用手工拟合方法并通过试凑来确定基底函数,存在设计效率低、拟合精度不高等不足。提出一种基于神经网络的飞行器控制方法,采用深度学习的方法实现对配平舵偏角、气动偏导数等气动参数的自动建模和计算,设计及仿真结果表明,提出的方法可大幅提升飞行器控制模型的通用性和设计效率。
The atmosphere vehicles control problem is a typical high uncertainty,serious nonlinear,strong coupling,fast time-varying system control problem. The design process of traditional feedforward & adaptive PD control law is faced with hypersurface fitting ploblem of trimming rudder angle and pneumatic partial derivative. It commonly uses manual fitting method and determines the critic function by trial and error. Thus,the traditional fitting method has low efficiency and fitting accuracy. This paper proposes a craft control method based on neural network,adopting the method of deep learning to automatically model the trimming rudder angle and pneumatic partial derivative. The result of design and simulation show that this method can greatly improve the versatility and design efficiency of aircraft control model.
The atmosphere vehicles control problem is a typical high uncertainty,serious nonlinear,strong coupling,fast time-varying system control problem. The design process of traditional feedforward & adaptive PD control law is faced with hypersurface fitting ploblem of trimming rudder angle and pneumatic partial derivative. It commonly uses manual fitting method and determines the critic function by trial and error. Thus,the traditional fitting method has low efficiency and fitting accuracy. This paper proposes a craft control method based on neural network,adopting the method of deep learning to automatically model the trimming rudder angle and pneumatic partial derivative. The result of design and simulation show that this method can greatly improve the versatility and design efficiency of aircraft control model.
引文
[1]郭宪,方勇纯.深入浅出强化学习:原理入门[M].北京:电子工业出版社,2018-1.
[2]赵冬斌,邵坤,朱圆恒,李栋,陈亚冉,王海涛,刘德荣,周彤,王成红.深度强化学习综述:兼论计算机围棋的发展[J].控制理论与应用,2016-6.
[3]陶九阳,吴琳,胡晓峰. AlphaGo技术原理分析及人工智能军事应用展望[J].指挥与控制学报,2016-6.
[4] Richard S Sutton,Andrew G Barto. Reinforcement Learning:An Introduction(2rd Edition)[M].. The MIT Press,2017-11.
[5]韩京清.自抗扰控制技术:估计补偿不确定因素的控制技术[M].北京:国防工业出版社,2008-9.
[6]崔富章.基于BP神经网络的空间飞行器姿态控制研究[D].重庆大学,2007. 04
[2]赵冬斌,邵坤,朱圆恒,李栋,陈亚冉,王海涛,刘德荣,周彤,王成红.深度强化学习综述:兼论计算机围棋的发展[J].控制理论与应用,2016-6.
[3]陶九阳,吴琳,胡晓峰. AlphaGo技术原理分析及人工智能军事应用展望[J].指挥与控制学报,2016-6.
[4] Richard S Sutton,Andrew G Barto. Reinforcement Learning:An Introduction(2rd Edition)[M].. The MIT Press,2017-11.
[5]韩京清.自抗扰控制技术:估计补偿不确定因素的控制技术[M].北京:国防工业出版社,2008-9.
[6]崔富章.基于BP神经网络的空间飞行器姿态控制研究[D].重庆大学,2007. 04