机动再入飞行器神经网络闭路初制导研究  

Initial Guidance Law of Closed-form Neural Network for Maneuverable Reentry Vehicle

作　　者：乔洋[1] 陈刚[1] 徐敏[1] 陈士橹[1] 

机构地区：[1]西北工业大学航天学院,西安710072

出　　处：《弹道学报》2007年第2期1-4,共4页Journal of Ballistics

基　　金：航天创新基金项目

摘　　要：解决有限推力模型的再入飞行器初制导问题,利用传统开环控制的关机方程达不到要求的精度.提出了一种前馈反馈神经网络闭路初制导律.利用遗传算法生成能量最省轨迹,作为神经网络训练样本,利用遗传算法离线训练前馈三层BP网络,构成前馈神经网络控制器.状态反馈控制器采用在线神经PID控制器.仿真结果表明,该闭路初制导能够完成有限推力情况下的再入初制导任务,对初始位置误差、推力方向偏差和测量误差不敏感,具有一定的鲁棒性和适应性.For the finite propulsion reentry guidance, the conventional closed-thruster equation based on open-loop control loses its way. A closed-form initial guidance law based on feed forward and feedback neural network control for finite propulsive reentry spacecraft was presented. Minimum energy trajectories were generated by genetic algorithm, which were used as training samples of three-tier feed forward BP neural network controller. A neural PID controller acts as a state feedback controller compensating the control errors of feed forward controller. Simulation results indicate that the closed-form guidance law with robustness is fit to the finite propulsive reentry mission and insensitive to the initial position errors, propulsion direction errors and observation errors.

关 键 词：神经网络 初制导律 遗传算法 最优轨迹 

分 类 号：V412.3[航空宇航科学与技术—航空宇航推进理论与工程]