复杂样本分类的GA-RBFNN方法

本文以提高径向基函数神经网络(RBFNN)分类能力为出发点,结合遗传算法(GA)群体并行搜索能力,提出了一种有效的GA-RBFNN学习算法.该算法在传统衰减聚类算法确定网络初始结构的基础上,加入控制向量,设计了包含整个网络隐节点结构和径基宽度的矩阵式混合编码方式,以及相应的遗传操作算子.网络权值由伪逆法求解确定.经Iris、WINES和Glass数据集的仿真实验验证,该算法快速有效,具有较强的复杂样本分类能力.

GA-RBFNN learning algorithm for complex classifications

This paper proposes an adaptive learning algorithm,GA_RBFNN,to build a RBF neural network(RBFNN) model.The algorithm utilizes GA's parallel_search ability to improve the classification accuracy of the RBFNN.Firstly,the initial network hidden structure of a RBFNN model is determined by the traditional decayed_radius clustering algorithm.Then the hidden centers of a RBFNN are modified by a specially designed GA,which is based on the matrix-form mixed encoding scheme with a control vector for regulating the structure of a RBFNN,and the new genetic operators are presented correspondingly.The pseudo-inverse algorithm is adopted to train the weights between the hidden layer and the output layer.Finally,experiments are implemented on datasets as Iris,WINES,and Glass,which shows that the proposed algorithm has higher classification ability compared with the conventional methods.