一种基于高斯核的RBF神经网络学习算法

RBF神经网络中心等参数确定得是否合理将直接影响到RBF神经网络的学习性能。通过有监督学习的方法来确定RBF神经网络的中心等参数是最一般化的方法。在这种方法中,参数的初始化问题是关键问题。文章在分析RBF神经网络映射性能的基础上,提出了中心等参数初始化的一种方法,并借助于梯度下降法给出了RBF神经网络的学习算法。多种实例表明,所给出的学习算法是有效的。该研究为RBF神经网络的广泛应用提供了一定的技术保障。

A Learning Algorithm of RBF Neural Networks Based on Gaussian Kernel Function

Whether the centers parameters etc.of radial basis function(RBF)neural networks are determined reasonably or not will directly influence approximation capability of RBF neural networks.It is frequently used algorithm to define these parameters by supervised learning method,but the determination of their initial values are key problems in this learning algorithm.Based on analyzing the mapping properties of RBF neural networks,a method is advanced for deter-mining the initial values of centers and other parameters.Meanwhile,with the help of gradient method,a learning algo-rithm is put forward for RBF neural networks based on Gaussian kernel function.Many experiments show that the learn-ing algorithm is very effective,so some foundations are laid for employing RBF neural networks in a lot of fields.