一种基于粗糙K-均值的椭球基函数神经网络

改进了一种椭球基函数神经网络,它与经典椭球单元神经网络的结构不同,而与径向基函数神经网络结构类似,即它有一个隐含层,并且隐层单元采用椭球基函数,区别于RBF网络的高斯函数。本文采用粗糙K-均值方法求取椭球函数的中心,并给出了该方法中确定初始阈值的步骤。这种改进方法不但使对输入空间的划分局部作用,而且划分区域封闭有界。因此,改进的神经网络具有较好的函数逼近能力和模式识别能力。仿真实验验证了该椭球基函数神经网络的正确性和有效性。

Ellipsoidal Basis Functional Neural Network Based on Rough K-Means

An ellipsoidal basis functional(EBF)neural network based on rough K-means is proposed.The structure of the netwok is similar to the radial basis functional(RBF) neural network rather than the conventional ellipsoidal unit one.In the new network,ellipsoidal unit functions are used in the hidden layer,and weights between hidden and output nodes are all connected.A method of rough K-means is used to obtain centers of EBF,and the way for deciding the threshold is given.The new neural network can make the partition of input space locally,and make it limitary and bounded.So the network has the capability of function approximation and pattern recognition.Finally,several examples are given.Simulation results show that these methods are correct and effective.