摘要
本文提出了高阶序贯非线性增广Ｋａｌｍａｎ滤波（ＳＥＫＦ），并将其应用于多层前馈感知器（ＭＬＰｓ）的学习问题．文中给出了ＭＬＰｓ的ＳＥＫＦ算法，得到了与ＢＰ算法类似的正向与反向传播过程，并且详细地推导了核心的量测Ｊａｃｏｂｉａｎ矩阵．结合一非线性正弦函数，ＤＥＫＦ和ＳＥＫＦ的仿真结果被进一步给出．
In this paper a higher order sequential nonlinear extended Kalman filter(SEKF),based on the DEKF algorithm given by ref.[4], is proposed and applied to the learning problem of MLPs The simulation result is shown that SEKF is superior to DEKF in the filtering accuracy and the required learning number except the slightly increased computational complexity.
出处
《控制理论与应用》
EI
CAS
CSCD
北大核心
1994年第3期381-384,共4页
Control Theory & Applications
关键词
前馈感知器
学习算法
非线性
multilayer feedforward perceptron
learning algorithm
nonlinear Kalman filtering
higher order sequential estimation
