一种带非线性扩展的前向神经网络模型及其学习算法

A Multilayer Feedforward Network Model of Inputs with Nonlinear Function and Its Algorithm

通过提出一种带非线性扩展的前向神经网络模型，分析了 Ｇ Ｇ Ｂ Ｐ 算法的收敛性，总结出此种算法的动态学习率．仿真结果表明：此神经网络模型更适合于处理多输入、多输出的问题，在这方面其收敛速度、逼近非线性函数的能力比函数型连接网络和前向网络都优越．采用动态学习率不仅可以保证网络的收敛性，而且可以使误差下降接近最快．

This paper develops a multilayer feedforward network model of inputs with nonlinear function, analyzes the convergence of the GGBP algorithm, generalizes the dynamical learning rate of the GGBP algorithm. The simulation results show : This network model much suitably deals with the problem of many inputs and many outputs; in this aspect, its convergence and capability of approximating the nonlinear function is better than the functional connection network and feedforward network. The dynamical learning rate not only ensures the convergence but also makes the error decrease nearly most rapidly.