RBF网的动态设计方法

Dynamic method for designing RBF neural networks

提出一种RBF网的动态设计算法 (DYNRBF方法 ) ,该算法有效地融合了ROLS算法和RAN网络的优点 ,不仅能动态调节RBF网的隐节点数 ,还能使网络的数据中心自适应变化 .该方法所设计的RBF网不仅具有较好的泛化能力 ,当训练样本集变化时也具有好的鲁棒性 .

A dynamic designing method to improve RBF nets' generalization ability, which is called DYNRBF method, is proposed in this paper. The method combines the advantage of ROLS algorithm and RAN net, so the number and position of data centers of RBF net are both adapted during learning progress. RBF nets designed with DYNRBF method can not only generalize well, but remain good performance when train set is changed.