基于F代表点的高斯过程矩匹配滤波  

F Discrepancy Rep-points Based Gaussian Process Moment Matching Filter

作　　者：姚智颖[1] 刘冬[2] 

机构地区：[1]第二炮兵工程学院303教研室,西安710025 [2]西安测绘研究所,西安710034

出　　处：《宇航计测技术》2011年第5期70-75,共6页Journal of Astronautic Metrology and Measurement

摘　　要：提出了一种基于F偏差代表点的高斯过程矩匹配滤波方法。该方法通过高斯过程学习系统的过程函数和观测函数,在高斯矩匹配滤波的框架下利用F偏差代表点进行数值计算。在单变量非线性模型及静基座初始对准中进行的仿真实验表明,该方法的精度不低于解析计算的高斯过程假设密度滤波(GP-ADF),远高于高斯过程无迹卡尔曼滤波(GP-UKF)。To solve the filtering problem with unknown system models, a F-discrepancy rep-pomts based Gaussian process moment matching filter is advanced in this paper. This filter learns the transition and observation models by Gaussian processes and then computes numerically in the structure of Gaussian moment matching filtering with F-discrepancy rep-points. Simulations in both single variants nonlinear sys- tem and the initial alignment problem have show that the filtering precision of this F-discrepancy rep-points based Gaussian process filter is not lower than GP-ADF and much better than GP-UKF.

关 键 词：F偏差代表点 高斯过程回归 贝叶斯估计 矩匹配滤波 

分 类 号：TN713[电子电信—电路与系统]