多层神经网络的快速BP算法

Quick BP Algorithm Used in Multilayer Neural Network

首先分析了BP算法中的误差函数对收敛精度的影响,给出了综合考虑绝对误差和相对误差的误差函数,其次利用优化方法中的共轭梯度算法来计算学习过程中的误差下降方向,并根据变尺度公式对该方向进行修正,从而得到快速BP算法。经过对实际算例的模拟,结果表明本文给出的快速BP算法是一种适用于多层神经网络的、性能忧良的学习算法。

In this paper, it is analyzed that the inflection of error function upon convergence precision in the BP Algorithm, and it is also given that the error function that considers comprehensively the absolute and relative errors. At the same time, the Conjugated Gradient Method is used to calculate the error descent direction and modifies the direction with the Variable Metric Algorithms, so the Quick BP Algorithm is given. The examples show that the Quick BP Algorithm has good capability and is applicable to the multilayer neural network.