多层前馈神经网络研究——单位BP算法

Research on Multilayer Feedforward Neural Networks ——A Unitary Backpropagation Algorithm

分析了引起标准 BP算法收敛速度慢的原因 ,以及传统改进方法的不足之处 ,探讨了解决的途径。为了提高 BP算法的收敛速度 ,定义并引入了基量函数的概念 ,并将其运用到 BP算法中 ,给出了一种高效的单位 BP算法。仿真和实例结构均表明该算法能够较好地克服标准 BP算法收敛速度慢的缺点 。

The reasons for the slowness in convergence of standard backpropagation algorithm and the imperfection of conventional improved algorithms have been fully analyzed. In order to improve the convergence rate of multilayer feedforward neural networks, a new highly efficent unitary backpropagation algorithm based on the unitary function is proposed. Numerical simulation and experimental results show that the algorithm can greatly increase the convergence rate and highly improve their imminent accurary.