一种新的在线自适应混合RBF网络学习算法  

A Novel Online Adaptive Mixed Learning Algorithm of the Radial Basis Function Neural Network

作　　者：翟华伟[1] 崔立成[1] 张维石[1] 

机构地区：[1]大连海事大学信息科学技术学院,辽宁大连116026

出　　处：《小型微型计算机系统》2014年第12期2713-2716,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61272172)资助;大连市基金项目(2013J21DW008)资助;青年教师基金项目(3132013042)资助

摘　　要：针对RBF网络序贯学习算法参数多、计算复杂等问题,深入分析RBF网络隐节点贡献度计算方法,提出基于主成分和周期性的贡献度计算方法,改进RBF网络GAP学习算法,细化算法中增加、删除和替换隐节点的条件,控制隐节点数量,自适应调整RBF网络结构.实验结果表明,相比传统RBF网学习算法,该算法在可靠性和泛化能力上都有显著提高.The sequential learning algorithms of the RBF network exist some problems, such as too many parameters, computing complex and so on. So, deeply analyzing computing methods of the importance of hidden nodes, the new methods based on principle components and periodicity are proposed, and it leads to reduce computation complexity, detail the preconditions of adding, deleting and updating hidden nodes. It introduces the overlap factor, strictly controls the amount of hidden nodes, self-adaptively adjusts the structure of the RBF network. After the tests, compared with other algorithms,it is superior in reliability and generalization capability.

关 键 词：RBF网络 隐节点贡献度 GAP学习算法 重叠因子 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]