基于敏捷性导弹逆动态的神经网络控制方法与仿真研究

给出了一种基于敏捷性导弹逆动态的神经网络控制方案。该方案由两个神经网络组成：第一个神经网络（NNI）用来离线的学习整个飞行包线内导弹动态特性的逆特性，以实现系统的线性化；由于敏捷性导弹在大迎角状态下具有高度的非线性特性和气动参数突变等未建模动态，因此引入第二个神经网络（NN2）来在线的补偿NNI的逆误差。在线学习的权值调整由Lyapunov理论得出，保证了闭环系统的稳定性。该控制方案对参数变化及未建模动态等具有良好的鲁棒性。将其应用于敏捷性导弹的控制中，数字仿真结果表明该控制方案有效。

Neural-Network Control Based on Agile Missiles' Inversion Dynamics

Neural-network control architecture based on agile missiles?inversion dynamics is presented. It consists of two neural networks. The first neural network is used to represent the nonlinear inverse transformation. It is trained off-line using a mathematical model, which provides an approximate inversion that can accommodate the total flight envelope. The second neural network capable of on-line learning is required to compensate for inversion error, which may arise from nonlinear dynamics, approximate inversion, or sudden changed in aircraft dynamics. A stable weights adjusting rule for the on-line neural network is derived from Lyapunov theory, thus assuring the stability of the closed-loop system. A benefit is that the control system tends to be more robust. Apply the control system to an agile missile, the digital simulation results show its effectiveness.