基于BP神经网络的自主定轨自适应Kalman滤波算法  

An Adaptive Kalman Filtering Algorithm for Autonomous Orbit Determination Based-on BP Neural Network

作　　者：尚琳[1,2] 刘国华[1] 张锐[1,2] 李国通[1,2] 

机构地区：[1]上海微小卫星工程中心,上海200050 [2]中科院上海微系统与信息技术研究所,上海200050

出　　处：《宇航学报》2013年第7期926-931,共6页Journal of Astronautics

基　　金：上海市科学技术委员会课题(10DZ2291700);上海市自然科学基金(11ZR1443500)

摘　　要：针对Sage-Husa自适应滤波方法存在的窗函数开窗大小选择问题,提出一种基于BP神经网络学习估计系统协方差矩阵的自适应Kalman滤波算法。该算法以Kalman滤波预测残差向量作为网络输入,通过网络分段离线学习确定预测残差向量与预测残差协方差矩阵间的非线性关系,自适应地估计Kalman滤波系统协方差矩阵。将其应用到自主定轨系统,仿真结果表明利用本文算法自主定轨60天星座平均URE误差小于1.9米,且能够快速跟踪到系统噪声的突变,较Kalman滤波方法和Sage-Husa自适应滤波方法具有更好的性能。In this paper, an adaptive eovariance matrix estimation algorithm based-on BP neural network learning is proposed to solve the window size selection problem in Sage-Husa adaptive filtering way. The innovation vector derived from the Kalman filter （KF） is employed as the input to the BP neural network and the nonlinear function between the innovation vector and the innovation covariance matrix can be determined through learning of the network. The eovariance matrix estimation algorithm proposed in this paper is applied to the autonomous orbit determination system. The simulation results show that the mean constellation URE of autonomous orbit determination will be within 1.9 meters in 60 days and it has better performance than the Sage-Husa adaptive filtering in the estimation of the system eovariance matrix of the autonomous orbit determination algorithm.

关 键 词：BP神经网络 自主定轨 自适应Kalman滤波 用户测距误差 

分 类 号：V249.3[航空宇航科学与技术—飞行器设计]