一种新型复合神经网络模型

A New Combined Neural Network Model

本文首先详细地阐述了ＢＰ神经网络和ＣＭＡＣ神经网络各自的结构，原理以及算法。提出了一种ＢＰ神经网络与ＣＭＡＣ神经网络组合起来的新型复合神经网络模型，并利用误差逆向传播原理推导出复合网络的学习法。仿真实验结果表明，这种复合神经网络在保留了ＢＰ和ＣＭＡＣ各自特长的基础上，同时具有学习速度快。

This paper first discusses the principle of two typical classes of neural network models: BP and CMAC, their structures, learning algorithms and approximation abilities. A new kind of Combined Neural Network(CNN) which uses the output of a CMAC neural network as an additional input node of BP neural network is then introduced. The corresponding learning algorithm is also derived by back propagating the approximation error in the output layer through each hidden layer to the input nodes. Comparisons of convergence speed and generalization ability have been made among BP, CMAC and CNN. Simulations suggest that the CNN has the advantage of fast learnig speed and good generalization ability. Further investigations are under discussion to explore this new neural network model to real time applications.