Abstract: This paper discusses the reasons of BP algorithm to be dropped in local minimization and puts forward three methods to avoid the situation .The first is to set the initial values of weight ,the second is to set the number of hidden layers reasonably , the last is to use the perception to disassemble the BP neural network and pre-set initial weight to avoid the BP algorithm to be dropped in local minimization. ;;;

摘要： 在详细地讨论了算法陷入局部极小原因的基础上,对BP算法避免陷入局部极小提出了3种基本的方法。第一种是预先给定权初值。第二种是合理地设置网络隐含单元数目。第三种是采用感知器分解预置初值的方法避免BP算法陷入局部极小。

关键词: 神经元网络, 算法, 局部极小, 感知器BP

CLC Number: 

• TP183