Using wavelet transform to improve generalization capability of feed forward neural networks in monthly runoff prediction

Abstract

In the study presented, a hybrid model is proposed for monthly runoff prediction by using wavelet transform and feed forward neural networks. Discrete wavelet transform (DWT) and Levenberg-Marquardt optimization algorithm based feed forward neural networks (FFNN) are considered for the modeling study. The study region covers the basins of Medar River which is located at the Aegean region of Turkey. Meteorological data, which represent the study region, were decomposed into wavelet sub-time series by DWT. Ineffective sub-time series were eliminated by using Mallow Cp coefficient based all possible regression method to prevent collinearity. Then, effective sub-time series components constituted the new inputs of FFNN. Some favorite evaluation measures, that is, determination coefficient (R2), adjusted determination coefficient (Adj.R2), Nash-Sutcliffe efficiency coefficient (NS), root mean squared error (RMSE), weighted mean absolute percentage error (WMAPE), were employed to assess modeling performances. The results determined in study indicate that the DWT based FFNN models (DWT-FFNN) are successful tools to model the monthly runoff series and can give good prediction performances than conventional methods. Key words: Wavelet transform, feed forward neural networks, Levenberg-Marquardt algorithm, monthly runoff prediction.