改进ACC算法优化的RBF神经网络研究及其应用  

Improved ACC Algorithm to Optimize RBF Neural Network and its Application

作　　者：陈铁军[1] 彭皎龙[1] 杨阳[1] 

机构地区：[1]郑州大学电气工程学院,郑州450001

出　　处：《计算机测量与控制》2013年第5期1323-1326,共4页Computer Measurement &Control

基　　金：河南省杰出人才创新基金(521000100)

摘　　要：通过对径向基函数(RBF)神经网络K-means训练算法聚类结果对初始中心点选择的依赖性以及容易陷入局部最优解等问题的分析,结合蚁群聚类(ACC)算法对蚂蚁睡眠行为进行模拟,并借鉴AM模型的聚类思想,根据样本分布,利用适应度函数和激活概率自主地确定隐节点数目和基函数中心,进而提出一种改进蚁群聚类算法优化K-means的混合聚类算法,该算法解决了RBF网络结构和参数辨识的问题;通过人工数据集进行测试,并通过应用实例进行验证,结果表明,基于混合聚类算法的RBF神经网络具有显著的优越性和可行性。According to K--means algorithm training process of radical basis function （RBF） neural network, the question is analyzed that clustering results must depend on initial point selected and can be easy to fall into local optimal solution. Therefore, combining with that Ant colony clustering （ACC） algorithm simulates ants sleep behavior. Drawing lessons from clustering thought of AM model. Using of the fitness function and the activation probability adaptively adjust the hidden nodes and basis function center by sample distribution. Then a hybrid clustering algorithm of improved ant colony clustering algorithm to optimize K--means was proposed. The algorithm solves the question of the RBF network structure and the parameter identification. Through the artificial data sets for testing, and through the application of the example verification, the results show that the RBF neural network based on hybrid clustering algorithm has the obvious superiority and feasibility.

关 键 词：径向基神经网络 改进蚁群聚类算法 混合聚类 群体智能 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]