一种基于BP网络的改进

Relative Entropy of BP Network

BP神经网络存在其固有的缺陷:收敛速度慢、易陷入局部极小、网络结构难以确定等.本文针对BP网络学习速度慢的缺点.用熵函数作为误差函数来对BP算法进行改进,提高了收敛速度和稳定性.通过对标准BP网络和相对熵方法在不同学习速率上收敛速度的比较实验证明,相对熵BP网络在收敛速度和稳定性方面有着明显的优越性.

There are inherent shortcomings in BP neural network： slow convergence and easy to fall into local minimum and difficult to determine the network structure.In this paper,entropy function is used as error function to improve the convergence speed and stability.Compared with the standard BP network on the learning rate in different comparative examples of the convergence rate,relative entropy BP network is superior to any of the above.