基于广义径向基函数的神经网络分类预测

Classification and Prediction of Neural Network Based on Generalized Radial Basis Function

径向基函数网络是神经网络中一种广泛使用的设计方法。它把神经网络的设计看作是一个高维空间的曲线逼近问题。相对于其他的神经网络方法,径向基函数神经网络除了具有一般神经网络的优点,如多维非线性映射能力、泛化能力、并行信息处理能力等,还具有很强的聚类分析能力,学习算法简单方便等优点。针对一个实际分类问题,利用广义径向基函数网络的思想训练一个网络并实现对测试数据集的分类预测。本算法采用k-均值聚类算法训练广义径向基函数网络中心,使用奇异值分解计算输出层权值。对该网络的实现细节及待改进之处进行简要分析。实验表明广义径向基函数神经网络的思想具有很强的聚类分析能力,学习算法简单方便等优点。

Radial basis functions （RBF） is a widely used tool in the neural networks. It views the design of the neural networks as a curve approaching problem in the high-dimensional space. Besides the advantages of the general neural networks, such as multi-dimensional nonlinear mapping capabilities, generalization, parallel information processing capabilities, Gaussian radial basis functions has strong ability of cluster analysis and its learning algorithm is simple and convenient. In this paper use Gaussian radial basis functions to solve a classified problem. First train the network and then realize classifying the data in the test set. The algorithm uses k-means clustering algo-rithm to generalize the training centers of the RBF network and singular value decomposition algorithm to calculate the output values. Finally, the analysis of the algorithm is given. The experiment shows that Gaussian radial basis functions has strong ability of duster analysis and its learning algorithm is simple and convenient.