WD-RBF Model and its Application of Hydrologic Time Series Prediction

Abstract

Accurate prediction for hydrological time series is the precondition of water hazards prevention. A method of radial basis function network based on wavelet de-nosing (WD-RBF) was proposed according to the nonlinear problem and noise in hydrologic time series. Wavelet coefficients of each scale were calculated through wavelet transform; soft-threshold was used to eliminate error in series. Reconstructed series were predicted by RBF network. The simulation and prediction of WD-RBF model were compared with ARIMA and RBF network to show that wavelet de-nosing can identify and eliminate random errors in series effectively; RBF network can mine the nonlinear relationship in hydrologic time series. Examples show that WD-RBF model has superiority in accuracy compared with ARIMA and RBF network.