비선형, 비정상 시계열 예츨을 위한 RBF(Radial Basis Function) 신경회로망 구조

초록

In this paper, modified RBF(Radial Basis Function) neural network structure is suggested for the prediction of a time-series with non-linear, non-stationary characteristic. Original RBF neural network predicting time series by past outputs is to sense the trajectory of the time series and to react when there exists strong relation between the input and hidden neuron's center. But this response is highly sensitive to level and trend of time series. To prohibit these, hidden neurons are modified to react to the difference of input and multiplied by encrement(or decrement) for prediction. When the suggested structure is applied to prediction of Mackey-Glass chaotic time series, Lorenz equation, and rossler equation, improved performances are obtained.