基于无轨迹卡尔曼滤波神经网络的SINS初始对准方法  

Application of Neural Network Based on Unscented Kalman Filter in SINS Alignment

作　　者：刘育浩[1] 黄新生[1] 

机构地区：[1]国防科学技术大学机电工程与自动化学院,长沙410073

出　　处：《系统仿真学报》2009年第10期2858-2861,共4页Journal of System Simulation

基　　金：兵器项目预研基金(2020203)

摘　　要：针对扩展卡尔曼滤波算法的缺点,研究了基于无轨迹卡尔曼滤波的神经网络算法。建立了捷联惯导系统在大失准角下的误差方程和适用于初始对准的神经网络模型,研究了基于扩展卡尔曼滤波和无轨迹卡尔曼滤波的神经网络算法,将基于无轨迹卡尔曼滤波的神经网络学习算法应用于捷联惯导系统大失准角条件下的初始对准中。仿真结果表明,在大失准角条件下,用训练好的神经网络进行初始对准是可行的,且基于无轨迹卡尔曼滤波的神经网络算法可提高初始对准精度。In view of the drawbacks of extended Kalman fiter （EKF）, the neural network （NN） based on unscented Kalman filter （UKF） was introduced. The error equation of Strapdown Inertial Navigation System （SINS） under large heading uncertainty and the NN model suitable for SINS alignment were built up. The learning algorithms of NN based on EKF and UKF were studied, and the algorithm of NN based on UKF was applied to SINS initial alignment under large heading uncertainty. Simulation results show that trained NN is available in alignment under large heading uncertainty, and the NN based on UKF improves the accurate in estimate precision.

关 键 词：捷联惯导 初始对准 扩展卡尔曼滤波 无轨迹卡尔曼滤波 神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]