基于RBF网络的捷联惯导初始对准优化研究  

RBF Networks for Initial Alignment of Strap-down Inertial Navigation System

作　　者：贺娟[1] 崔平远[1] 陈阳舟[1] 李振龙[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100022

出　　处：《计算机仿真》2006年第4期30-32,49,共4页Computer Simulation

基　　金：863基金资助资项目课题(KL0202200201)

摘　　要：该文研究了捷联惯导在静基座下的初始对准问题,其中粗对准中采用磁罗盘和加速度计得到初始姿态角,精对准中采用具有最佳逼近性的径向基函数神经网络(RBF网络)修正姿态角。在RBF网络基函数中心选取时,基于样本分布特点,采用简单有效的均值法,同时为了增加了神经网络权值学习的鲁棒性和快速性,将H∞鲁棒滤波用于网络的权值调整中。仿真结果表明,采用这种优化学习的RBF神经网络进行初始对准比用传统的Kalman滤波更快速有效,且与Kalman滤波精度相当。This paper presents a method for the initial alignment of strap - down inertial navigation system on stationary base. Magnetism compass and accelerometer are used to get the initial pose angles for the coarse alignment, Then RBF neural network, with its radial function chosen by average method and weights adjusted by H∞ robust filter, is applied for changing the pose angles to ensure the robustness and real - time computing. Simulation results show that this method is efficient and rapid for initial alignment compared with traditional Kalman filter, and it can achieve the comparative precision of Kalman filter as well.

关 键 词：捷联惯导 初始对准 神经网络 鲁棒滤波 

分 类 号：V448.15[航空宇航科学与技术—飞行器设计]