基于ART与RBF网络的混合网络模型设计

Design of a Hybrid Network Model Based on Adaptive Resonance Theory(ART) and Radial Basis Function(RBF) Network

为了解决自适应共振理论(adaptive resonance theory,ART)网络对新的输入样本处理能力差,网络记忆能力差,径向基(radial basis function,RBF)网络选择径向基函数,确定隐节点数目困难的问题,设计了一种基于ART与RBF网络的混合网络模型。将ART网络的特点引入到RBF网络中,通过ART网络的识别与比较功能快速确定RBF网络最少的隐节点数目,同时通过ART理论中警戒门限的检验在线确定是否合并或删除隐节点。并且引入了异常数据修正方法和模糊预处理方法。通过MATLAB神经网络工具箱,对该混合网络进行仿真试验。结果表明:该方法能够有效地减少隐含在数据中的随机性,加快神经网络收敛速度,提高神经网络的建模精度。

In order to solve the shortcomings of the poor ability of processing new input samples with adaptive resonance theory（ART） network and the poor memory of the network,and the difficulties of selecting radial basis function（RBF） and determining the number of hidden nodes of a RBF network,a hybrid network based on ART and RBF network model is designed in this paper.The characteristics of ART network are introduced into RBF network,then the least number of hidden nodes of RBF network is quickly identified through the identification and comparison function of ART,and the online merger and deletion of hidden nodes are determined through the alert threshold of ART.Moreover,the methods for amendment of abnormal data and fuzzy preceding operation are used.The hybrid network is simulated through the neural network toolbox of MATLAB.The results show that the method can effectively reduce the implied randomness of data and speed up the convergence rate of the neural network,and the accuracy of neural network modeling is improved.In the forecasting of network,the hybrid network model is superior to the single RBF network model.