【作者】 冯斌；

【导师】 姜长生；

【作者基本信息】 南京航空航天大学 ， 模式识别与智能系统， 2003， 硕士

【摘要】 本文就神经网络自适应控制在直升机飞行控制系统中的应用进行了深入的研究和探索，主要的内容包括直升机飞行控制系统的动态逆设计、小波神经网络对动态逆误差的补偿、直升机机动飞行的仿真等方面的内容。 由于直升机存在着较强的耦合性、不稳定性、呈现出较强的非线性，本文首先建立了直升机非线性运动方程，并对自由飞机的开环特性进行了研究，分析了运动耦合和操纵耦合对直升机各状态的影响，揭露直升机各通道之间的耦合的原因，为下面的控制系统的设计提供一定的物理概念。 本文将直升机姿态角通道和总距通道分别进行设计。在姿态角通道的设计中，采用了动态逆思想，按照时标分离的原则将直升机飞控系统划分为快慢回路，并分别采用动态逆的方法进行设计、仿真，使直升机的三个姿态角通道得到解耦。 针对由于模型不精确所引起的动态逆控制器的逆误差，采用小波神经网络实现补偿。本文对小波理论进行了讨论，给出了小波函数的一般构造。针对动态逆产生的逆误差，本文首先对基于BP小波神经网络进行了研究，给出了学习算法和权值调整律，设计了神经网络控制器；然后对基于李亚普诺夫稳定的小波神经网络进行了研究，对神经网络自适应控制律进行了详细的分析和讨论，设计了神经网络控制器。本文对这两种方法设计的神经网络控制器分别进行了仿真，表明它们能够很好的补偿动态逆产生的误差。 本文最后设计了直升机模糊PD轨迹跟踪控制器，与姿态角控制器相结合，完成对整个飞行控制系统的设计。并对两种典型的机动飞行(垂直平面中的急跃升／急跃降、水平平面中的S形机动)进行了数字仿真，结果表明本文设计的控制系统提高了直升机的机动性能。更多还原

【Abstract】 The design of neural-networks-based adaptive flight control system for helicopter is studied in this dissertation, which involving several aspects about inverse dynamics design for helicopter flight control system, compensating inverse error by wavelet neural-networks, maneuver flight simulation for helicopter.Because model of helicopter is a highly coupled unstable nonlinear system, a group of nonlinear motion equations are established firstly, and then analyzed the infection of the couple of motion and the couple of control of the free helicopter. This work supplied the physics concept for the design of the flight control system.In this dissertation the control loop of attitude angle is designed with dynamic inversion scheme. The controller is designed at two time scales. And the simulation shows the uncoupling of three attitude angle loop is achieved with dynamic inversion scheme.Adaptive wavelet neural network is introduced to compensate the inversion error resulted from inaccurate model of helicopter. The dissertation discussed wavelet theory and gave the formation of the wavelet function. In order to compensate the inversion error, the dissertation studied the BP based wavelet neural network firstly, and gave the learning rate; then studied the Lyapunov based wavelet neural network, and the adaptive control law of on-line neural network is discussed at great length. Two neural network controllers designed for both methods. The simulations show the both controller could compensate inversion error effectively.Finally, the dissertation designed the fuzzy PD controller of trajectory tracking, which is combined with attitude angle controller, so that the whole flight controller is completed. The simulation of two typical maneuver (BopUp-BopDown and S maneuver flight) shows that the maneuverable of attack helicopter is improved by the control system.更多还原