摘要
在采用Kalman滤波进行捷联惯导精对准时,当模型存在误差或系统噪声不能反映实际噪声时,会降低滤波精度甚至导致滤波发散。针对这个问题,提出基于Elman神经网络和Kalman滤波的捷联惯导精对准方法。首先对已知噪声统计特性的系统进行Kalman滤波,将稳定可靠的状态估值作为网络期望输出用来训练Elman网络;然后再用训练好的网络对未知噪声统计特性系统进行状态估计。利用仿真数据对该算法进行验证,结果表明,该算法能够克服Kalman滤波精对准的缺陷,提高对准精度,尤其是航向角的精度。
In SINS refined initial alignment using Kalman filtering, when an inaccurate model is used or the systematic covariance matrix doesn't indicate the actual noise, it will degrade the filtering accuracy or even lead it to divergence. In order to solve this problem, a new method based on Elman neural network and Kalman filtering was presented. First, the reliable state estimation of Kalman filtering for the known system was taken to train the Elman neural network. Then the trained neural network was applied to estimate the state parameters for the unknown system. By the simulating data, it is verified that this new algorithm can overcome the shortcomings of Kalman filtering in SINS refined initial alignment and improve the alignment accuracy, especially the yaw accuracy.
出处
《中国惯性技术学报》
EI
CSCD
2006年第5期9-13,共5页
Journal of Chinese Inertial Technology
基金
国家自然科学基金项目(40274002和40474001)
作者简介
吴富梅(1981-),女,硕士,主要从事动态大地测量数据处理.电子邮箱:wfm8431812@163.com
