摘要
新一代空间望远镜采用分块式主镜方案,以多个旁瓣子镜背后布置的若干致动器为控制手段,对主镜面形进行高精度控制,是一个复杂的控制系统。应用BP神经网络的方法建立了以致动器作用力为输入、镜面形变的Zernike多项式拟合系数为输出的镜面形变模型,并利用镜面形变的有限元分析的大量数据,对BP模型进行离线训练和模型预测效果的检验。仿真结果表明所建BP模型接近有限元分析的精度,并可以满足空间望远镜在线控制的实时性要求。
The segmented primary mirror of next generation space-based telescope was controlled by a number of actuators to achieve a required surface shape. The back-propagation (BP) neural network was applied to modeling the primary mirror surface with inputs of actuator applied force and outputs of Zernike polynomial coefficient. By using primary mirror finite element analysis data, BP model was trained offline and its predictive precision was verified. The simulation results show that the precision of BP neural network model is approximate to the finite element analysis, and it satisfies the requirement of space-based telescope real-time control.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2006年第z2期788-790,793,共4页
Journal of System Simulation
关键词
分块式主镜
面形控制
有限元分析
BP神经网络
segmented primary mirror
surface control
finite element analysis
BP neural network