摘要
本文针对BP算法存在收敛速度慢的缺点，提出一种基于网络动态训练误差变化率自动校正学习步长和冲量因子的自适应反向传播算法.异或问题、非线性系统和参数波动系统辨识的结果表明.
In order to overcome slow convergence rate of the standard BP algorithm, this paper presents an adaptive backpropagation algorithm which can update learning rate and birr factor automatically based on dynamical training error rate of change. Simulation results of the XOR problem, nonlinear system and parameter wing system show much faster convergence rate can be obtained by this algorithm.
出处
《电子学报》
EI
CAS
CSCD
北大核心
1998年第10期141-144,共4页
Acta Electronica Sinica
基金
山西省回国留学人员基金
关键词
神经网络
反向传播算法
XOR问题
Neural networks,Adaptive backpropagation algorithm, XOR problem, Nonlinear system, Parameter varying system
