神经网络在地-地导弹惯导系统初始对准中的应用

以地-地导弹的惯导系统为研究对象,分析了传统方法在惯导系统初始对准方面的缺陷.针对惯导系统的非线性及实时性等方面的要求,考虑到神经网络所具有的函数逼近性能,扩展Kalman滤波(EKF)所具有的最优估计性能的特点,提出了基于扩展Kalman滤波的神经网络应用技术.应用扩展Kalman滤波对多层神经网络的非线性离散时间系统进行算法训练,在获得的所有观测数据中找到状态(权值)的最小方差估计.在假定的地理坐标系下,对地-地导弹的惯导系统地面自对准的非线性状态方程,应用Matlab对基于EKF的神经网络方法和传统的Kalman滤波方法进行了仿真,对仿真的结果进行了对比分析.

Neural network based INS initial alignment for ground-to-ground missiles

We studied the Inertial Navigation System(INS) of ground-to-ground missile,and analyzed the drawbacks of traditional methods in initial alignment of INS.INS has the requests in non-linear and real-time performance.Considering the function-approximation performance of neural network,and the optimization-estimation performance of Extending Kalman Filtering(EKF),we put forward the technique of neural network based on EKF.EKF was used for arithmetic training of non-linear discrete time system of multi-level neural network,the minimum variance estimation of weighted values was found in all the observed data.To study the nonlinear state equation of auto-alignment system of INS in ground-to-ground missile,we carried out simulations by use of Matlab for EKF based neural-network method and traditional Kalman filtering method under supposed geography coordinates.The result of simulations was analyzed in this paper.