A highly nonlinear artificial neural network(ANN ) model is introducedas analternative approach for river flow forecasting of the Hengshi station

the control stationof the largest key project of water and electricity in Guangdong province。 Data for modelcalibration and test include five year hourly and one year daily mean discharge·Themodel can efficiently simulate the nonlinear process of a real hydrological system

Com-parisons are made between the ANN model and the existing system analysis models

e.g.CAR

RWTL and AR models. Totally l49 models are identified in this study

in allcases

the ANN model is superior to the system analysis models in the sense of increas-ing the forecasting lead time and improving the forecasting quality at the mean time。