Forecasting water demand using back propagation networks in the operation of reservoirs in the citarum cascade, west java, indonesia

Abstract

This study investigates the use of Neural Networks (NN) as a potential means of more accurately forecasting water demand in the Citarum River basin cascade. Neural Networks have the ability to recognise nonlinear patterns when sufficiently trained with historical data. The study constructs a NN model of the cascade, based on Back Propagation Networks (BPN). Data representing physical characteristics and meteorological conditions in the Citarum River basin from 1989 through 1995 were used to train the BPN. Nonlinear activation functions (sigmoid, tangent, and gaussian functions) and hidden layers in the BPN were chosen for the study.