BP神经网络权值初始值与收敛性问题研究

RESEARCH OF THE RELATION BETWEEN THE INITIAL WEIGHT AND CONVERGENCE OF BP NEURAL NETWORK

BP神经网络的收敛性问题是一直受到广泛关注的问题。本文针对 BP网络在运算过程中陷入局部最小区、收敛速度慢的问题 ,从 BP算法的原理出发 ,讨论了权值初始值对网络训练速度的强烈影响 (仿真结果证明了这一点 ) 。

The problem of the convergence of BP neural network is proposed in this paper.In order to improve the convergence characteristics,we discuss the strong effect of the initial weight to the training speed based on the theory of the BP algorithm,which is proved by the results.