基于熵误差函数的BP算法及其应用

BP算法现在已成为目前最广泛的神经网络学习算法之一,但存在收敛速度较慢和学习不稳定的问题.为了加快收敛,用熵作为误差函数来对BP算法进行改进,在训练过程中加入动量项,并且对样本作归一化处理.通过对函数逼近和异或问题实例的仿真,与通常改进的BP算法所得的结果进行比较,仿真结果表明熵作为误差函数的改进BP算法比通常改进的BP算法有更好的收敛性和稳定性.最后,通过改进的BP神经网络实现对矩形波导匹配负载的结构设计,结果表明网络能很好地达到工程的要求.

A BP Algorithm Based on Entropy Error Function and Its Application

BP (back propagation) algorithm has become one of the most widely used neural network learning algorithms. However, it has the problems of slow convergence and unstability. In order to accelerate the convergence, BP algorithm is improved with entropy error function, moment term is added to the network, and input and output sample are scaled linearly. The paper presents the results of Function Approximation and XOR problem between the two networks, and the simulation results show that this algorithm has better convergence and stability. Finally, the matched load of rectangular waveguide is designed in the optimized network, and the result shows that this network can meet the requirement of the project.