Prediction by neural network and information criterion on daily inflow in a dry season

Abstract

This study aims at the prediction of daily inflow into a reservoir area by neural network model and information criterion (e. g., AIC) in a dry season. The calculation of network is carried out by the back propagation learning with a modified moment method. Comparison with this method and the usual linear regression method shows that neural network is effective in the reduction of error variance and in the increase of correlation between the estimation and observation of inflow data