摘要
提出了一类带有离散时间 FIR/ IIR滤波器的递归 RBF神经网络 ,用离散时间 FIR/ IIR滤波器代替通常的 RBF神经网络中的线性输出权值 ,以适用于离散动力学系统的辨识和控制以及混沌时间序列预测 .本文给出的学习算法简单 ,可以避免传统的递归算法的不稳定性 .将该类神经网络用于动力学系统的建模 ,收到很好的效果 .
A class of recurrent RBF neural networks with discrete time FIR/IIR filters are presented. The discrete time FIR/IIR filters in their output layers replace the output linear weights in general RBF neural networks in order to apply them to identification and control of discrete dynamic systems and prediction of chaotic time series. The learning algorithms are simple and can avoid unstability existing in conventional recurrent algorithms. Application of the new neural networks to modelling dynamic systems shows their superiority.
出处
《系统工程理论与实践》
EI
CSCD
北大核心
2001年第10期47-51,共5页
Systems Engineering-Theory & Practice
基金
国家自然科学基金 (60 0 740 0 8)
关键词
递归径向基函神经网络
学习算法
FIR滤波器
IIR滤波器
recurrent radial basis function
dynamic system
finite impulse response(FIR)
infinite impulse response(IIR)
filter
