针对巡逻导弹,讨论一种将神经网络与非线性动态逆相结合的自动驾驶仪设计方法。基于双时标假设,将弹体动力学分为慢变子系统和快变子系统,并分别进行动态逆设计。为克服模型不精确和参数摄动引起的逆误差对动态逆控制的不利影响,引入神经网络对逆误差进行在线补偿。推导了神经网络权值调整规律,并证明系统的闭环稳定性。通过仿真验证,采用神经网络能弥补动态逆控制的不足,提高控制系统的鲁棒性,是一种有效的非线性设计方法。

A nonlinear dynamic inversion control method based on neural network is applied for loiter missile autopilot design. Firstly, the dynamics of the missile are separated into a fast part and a slow part based on the two time-scale assumption. Then dynamic inversion control method is appliyed for both the two parts. The neural networks(NN) are adopted to compensate the dynamic inversion error casued by inaccuracy of model and system parameter perturbation. The adjustment arithmetic of NN weight is given and stability of the closed loop system is proved. Finally, the simulation results demonstrate that the method proposed overcomes the shortage of nonlinear dynamical inversion control scheme where a precise mathematic model is necessary and system robustness has improved. The effectiveness of the nonlinear control scheme is proved.