Multi Layer RBF Network’s Clustering Method And Its Applications

Single radial basis function network with a single hidden layer，is a three layersfeedforward neural network，it has been widely used in function approximation，nonlinear regression and time sequence and other fields， and have achieved goodapplication effect. The radial basis function network trains mainly focus on how to getfaster to get better clustering centers. This paper changes the perspective， presents aconception of multi-layer radial basis function network and the clustering trainingmethod which can put forward to the idea. the clustering training method is composedby Moody and Darken algorithm to evolve and get.Presenting multi-layer radial basisfunction network’s original intention is to improve the network for nonlinear functionapproximation accuracy， to improve its performance.This article first elaborated the multi-layer radial basis function network theory andconception， and then puts forward the specific realization of this network clusteringalgorithm， and the use of computer simulation experiments to prove its highprecision approximation of functions on the real ability， also has the flexibility toadapt to a variety of nonlinear function of nonlinear regression function.On this basis，we use the multi-layer radial basis function network for Logistic andMackey Glass chaotic time series prediction. Because of this network with highprecision approximation of real functions， prediction accuracy is significantly higherthan that of single radial basis function network， the network in the prediction ofchaotic time series on effectiveness.Finally，the paper put the multi-layer radial basis function network into the basedon the price range of financial volatility models. The results show that，by using theS&P500index near14years day highest and lowest prices obtains range，and then onthe price range of volatility modeling and prediction. Multi-layer radial basis functionnetwork financial volatility forecasting，can greatly improve the prediction accuracy，thus proving the radial basis function network in nonlinear time series analysis oneffectiveness.