一种新的多层前馈网络学习算法  

A Novel Learning Algorithm for Multilayer Feedforward Neural Networks

作　　者：黄远灿 孙圣和 

出　　处：《哈尔滨工业大学学报》1995年第5期72-76,共5页Journal of Harbin Institute of Technology

摘　　要：基于仅值解耦思想，本文提出一种新的多层前馈网络学习算法。该学习算法的收敛速度比ＢＰ学习算法快，且比并行递推预测误差学习算法有更好的数值稳定性。文中论述了它与别的学习算法之间关系，并选取神经网络对非线性系统建模的例子进行仿真，说明新算法的有效性。Based on the idea of weight decoupled, a novel learning algorithm is derived for multilayer feedforward neural networks. It has better convergence than the back propagation learning algorithm and better stability than the parallel recursive prediction error learning algorithm.The relationship between this algorithm and other exsiting learning algorithms is discussed.Example taken from the nonlinear system modelling using neural networks are simulated ic demonstrate the effectiveness of the novel algorithm.

关 键 词：多层前馈网络 学习算法 非线性系统 

分 类 号：TP11[自动化与计算机技术—控制理论与控制工程]