基于步长优化和共轭梯度法的改进BP算法

A Improved Algorithm Based Step Optimum and Conjugate Gradiend

本文针对传统ＢＰ算法存在的两个常见问题进行了讨论，提出了基于步长优化和共轭梯度法的改进ＢＰ算法。新的算法避免了目前由人为经验确定迭代步长的缺陷，并能有效克服局部极小值。这样可使网络加快收敛，稳定性变好。

This paper discussed two basic problems of conventional BP algorithm.A improved BP algorithm based on step optimum and conjugate gradient was put forword in this paper.The new algorithm avoided the local minimum and the shortage nowadays that the learning step is determined by human experience.So it Could speed up the convergence and enhance the stability of network.