Multi-stage RBF Network Orthogonal Least Squares And Regularized Least Squares Learning Algorithm

Radial basis function network is a kind of excellent performance,with high operation speedand extrapolation capability of artificial neural network,with a strong biological background,hasattracted a lot of attention,and is widely applied in many fields,it is in the economy,the bondmarket,commercial banks,exchange rate,market price index etc.When economists and financial experts forecast the economy, they found that the traditionallinear regression model prediction is difficult to achieve excellent results. With the research andapplication of nonlinear theory of rapid development tools, RBF network models are widely usedin the field of economic and financial forecasts, it can reaches a better prediction than thetraditional linear regression model.In this paper, we use multi-stage RBF network model to predict China’s grain output. At thebeginning of this paper, we start to introduce the research background of RBF neural networkfunction and significance, the development of the RBF network model, and studies in the field ofeconomic and financial applications; then described the structure and algorithms based of RBFnetwork, with emphasis on the orthogonal least squares and regularized least squares learningalgorithm, and we propose multi-stage orthogonal least-squares method and multi-stageregularized least squares method based on former algorithms, then we verified results bycomputer simulation experiments and found that the two new methods can provide betterapproximation of experimental data. Finally, not only by comparing the two predictions ofChina’s grain output between by using multi-stage RBF network model and the ordinaryregression model, but also by comparing with Wu Yuming, Xu Jianhua’s RBF network modelprediction, multi-stage RBF network model is better than linear regression model for muchhigher precision, and more accurate than the RBF network model prediction.