具有快速学习算法的神经网络及其在系统辨识中的应用

A Neural work with Fast Learning Algorithm and Its Application to System Identification

提出一种基于平行切线的前向神经网络快速学习算法 ,学习速率采用曲线拟合法进行一维搜索。新方法具有很快的收敛速度和良好的收敛精度 ,克服了标准 BP算法在神经网络的权值训练中收敛速度慢的缺点 。

In this paper, a fast learning algorithm of forward neural net work adopting parallel tangents is presented in which the learning rate is searched for in one dimension by curve fitting method. The new algorithm overcomes the shortcoming of low convergence rate in standard BP net work algorithm and thus has the advantage of fast convergence rate and high convergence accuracy. Its effectiveness is proved by the simulation results.