一种基于双曲函数的BP网络改进算法

An improved algorithm based on hyperbolic function of BP neural network

提出了一种新的BP(Back Propagation)网络权值调整规则算法。该算法以双曲函数为基础,将BP网络输出层节点的实际输出值与期望输出值的差作为双曲函数的输入,输出为权值的变化量。实例仿真表明,该算法简单,容易实现,加快了网络的学习速度,提高了学习效率。

An improved algorithm of BP（Back Propagation） neural network is proposed, and a new weight- adjusting rule is introduced in the paper. The algorithm is based on hyperbolic function whose input is the error between actual output value of the output layer node and the expected output value, and the output is a variation value of weight. The simulation of instances show that the algorithm is simple and easy to implement. Meanwhile, it quickens the study of the network, and improves the efficiency.