Abstract: In this paper, an adaptive radial base function neural network was presented. In this network, the SOＭ neural network, as a cluster network, performed unsupervised learning and weight vectors belonging to its output nodes were transmitted to the hidden nodes in the RBF networks as the centers of RBF activation function, as a result one to one correspondence relationship was realized between the output nodes in SOFM and the hidden nodes in RBF networks. The RBF network, as a basic network, performed the nonlinear mapping from input nodes to hidden nodes using Gauss function and linear mapping from hidden nodes to output nodes using supervised learning algorithm. The simulations’ results on the recognition of the set of English character were shown to prove the proposed networks had good performance.

摘要： 提出了一种新的结构自适应的径向基函数(RBF)神经网络模型。在该网络中，自组织映射(SOM)神经网络作为聚类网络，采用无监督学习算法对输入样本进行自组织分类，并将分类中心及其对应的权值向量传递给RBF神经网络，作为径向基函数的中心和相应的权值向量；RBF神经网络作为基础网络，采用高斯函数实现输入层到隐层的非线性映射，输出层则采用有监督学习算法训练网络的权值，从而实现输入层到输出层的非线性映射。通过对字母数据集进行仿真，表明该网络具有较好的性能。