摘要
针对递归神经网络传统BP学习算法收敛慢的缺陷,将Levenberg-Marquardt(LM)算法引入到对角递归神经网络权值的训练,这种算法提供了快速性与收敛性之间的一个折衷。仿真结果表明,该算法比传统BP算法具有更快的收敛速度,用于非线性动态系统的建模是有效的。
A new recursive learning algorithm of diagonal recurrent neural network is proposed by introducing Levenberg-Marquardt （LM） algorithm. And this new algorithm provides the compromise of the fast speed and the convergence. Simulation results show that the proposed algorithm converges faster than the traditional recurrent BP algorithm, and can be used effectively in the nonlinear dynamic system modeling.
出处
《机械工程与自动化》
2008年第6期68-70,共3页
Mechanical Engineering & Automation
关键词
对角递归神经网络
LM算法
非线性动态系统
系统建模
diagonal recurrent neural network
LM algorithm
nonlinear dynamic system
system modeling
