摘要
RAN(Resource Allocating Network)是一种基于径向基函数的单隐含层神经网络 ,开始时无隐含节点 ,可根据新颖性准则来决定是否增加隐含节点 ,故有效避免了网络结构 (隐含节点个数 )和初始网络参数难以选取的缺点。本文首先利用 RAN得到网络的初始结构及参数 ,然后利用扩展的卡尔曼滤波器 EKF(Extended Kalman Filter)算法对网络参数进行调整 ,这相当于在粗调的基础上对网络参数进行细调。该网络具有学习速度快、精确度高、结构紧凑的优点 。
RAN(Resource Allocating Network) is a single hidden layer network, its hidden units are Gaussian radial basis function. RAN starts with no hidden units and grows by allocating hidden units based on novel criterion. First, we arrive at a initial network by RAN. Second, we tuned the parameters of the neural network by EKF(Extended Kalman Filter) algorithm. The advantages of RAN include quick learning, high accuracy and compact forms. It also can be applied in real time system.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2001年第1期13-16,共4页
Chinese Journal of Scientific Instrument
基金
国家 8 63青年基金资助项目!No:863- 82 0 - Q- 6