面向高超声速飞行器的 新型复合神经网络预测控制方法
A Novel Composite Model Predictive Control Method Based on Neural Network for Hypersonic Vehicles

针对高超声速飞行器(HV)受外界持续干扰影响以及非对称约束限制的问题,提出了一种基于神经网络的新型复合预测控制方法。首先,为了实现HV对攻角和控制信号这2类非对称约束的限制,利用多面体不变集处理非对称约束的能力,基于HV多个速度和高度的离散点,构建多面体不变集序列,并设计HV标称纵向模型(无持续干扰)的标称预测控制器;然后,基于神经网络,设计自适应神经网络干扰观测器估计HV的速度、攻角、俯仰角速率3个回路的干扰,并获得补偿控制器,有效抑制持续干扰。仿真结果表明:在HV受到外界持续干扰以及非对称约束限制的情况下,该方法可保证高度跟踪误差和速度跟踪误差分别收敛到10－4 km和10－4 km/s的范围内,而无补偿控制方法的跟踪误差则会大幅度振荡。
A novel composite model predictive control method based on neural network is proposed for a constrained hypersonic vehicle (HV) in the presence of external persistent disturbances and restriction of asymmetric constraints. Firstly, a stable polyhedral invariant set is constructed based on multiple discrete points of velocity and height of HV, and the ability of the set to deal with asymmetric constraint is used to limit asymmetric constraints of control signals and angle of attack. A nominal predictive controller for the nominal longitudinal theoretical model of HV without external persistent disturbances is designed. Secondly, an adaptive neural network disturbance observer is designed, and the observer is used to estimate the disturbances in loops of velocity, angle of attack and pitch rate. An auxiliary compensation controller of the nominal predictive controller is appended to effectively attenuate the influences of external persistent disturbances. Simulation results show that when HV is affected by persistent disturbances and limitation of asymmetric constraints, the proposed method guarantees the tracking errors for altitude and velocity to converge into sufficiently small ranges of 10－4 km and 10－4 km/s, respectively, while the tracking errors from the uncompensated control methods are oscillating in large amplitude