神经网络BP算法的误差分级迭代法

Error grade iterative method of BP neural networks

本文结合某一工程实例 ,对BP算法进行了改进 ,提出了误差分级迭代法 .通过实例分析 ,该方法确能提高收敛速度 ,克服初始权值的影响 ,同时 ,学习样本次序对其影响也不大 .因此 ,该方法能有效地改善BP网络的性能 .最后 。

Through an engineering example, a betterment algorithm of BP network, named “the error grade iterative method”, is innovated. The analysis of the data from the engineering example shows that the new method can speed up the convergence, overcome the influence of the first weights, and reduce the influence of the sequence of study samples. By using the betterment method, the performance of BP neural networks can be improved. At last, the working theory of the error grade iterative method is analyzed.