一种复合型前馈神经网络结构

ＢＰ网络是一全局逼近的网络，通常，在对样本数据的选取要求上，它不如ＲＢＦ网络高，即ＢＰ网络在训练样本数据的代表性不强时也可表现出对非线性函数较强的逼近性能，鉴于此，作者提出了一种复合型前馈神经网络结构在此结构中，笔者采用了ＢＰ神经网络对训练数据进行预处理而得到径向基函数（ＲＢＦ）网络的初始中心矢量点集的方法该方法使得ＲＢＦ网络中心矢量点集由传统的随机确定改为对它的优化选取，由它确定的ＲＢＦ网络的中心矢量具有相当的柔性，从而增强了径向基函数（ＲＢＦ）神经网络整体的泛化性及鲁棒性，最终使得该复合型神经网络具有了很好的精度和泛化性仿真结果表明了本文所提网络的有效性

A Mixture Structure of Feed-Forward Neural Network

The BP neural network is an overall approximation network. The requriement of sample data for BP network is less strict than that of RBF neural network. When the sample data is not very typical of all the model of the object, the BP network can appear good approximation ability to non-linear function. According to this character, the authors raise a mixture neural network structure. In this structure, we use BP neural network to select the center of RBF neural network. This method lets the center vector of RBF network to be selected optimized. As a result, the simulation of using this mixture network appears good adaptability and robustness.