【摘要】 状态相关黎卡提方程(SDRE)需要在每个采样点求解矩阵Riccati方程,计算成本非常高,不利于实时控制器的实现。将SDRE方法与鲁棒H∞方法相结合,针对导弹纵向自动驾驶仪设计SDREH∞控制器,利用一种多层回归神经网络结构来进行在线求解代数Riccati方程,从而获得控制器增益。神经网络学习方法采用常规梯度法。仿真结果表明导弹纵向跟踪控制性能好;神经网络在线求解代数Riccati方程有效降低计算成本,且利用硬件实现神经网络求解器。更多还原

【Abstract】 The difficulty lies in the implementation of SDRE(State Dependent Riccati Equation) controller is computational cost on solving matrix Riccati Equation in each sample point.A SDRE H∞ controller is designed for missile longitudinal autopilot.A multilayer neural network scheme is used to solve algebraic Riccati equation,resulting from robust H∞ controller design.The simulation results reveal effectiveness of the tracking controllers and efficiency of neural-net scheme for solving Riccati equation.It is convenient to realizing a neural-net solver by hardware.更多还原