According to the nonlinear characteristics of the daily grid electricity data of hydropower stations, the statedependent autoregressive (SDAR) model is used to describe this data. In this study, the Gaussian radial basis function (RBF) networks are used to approximate the functional coefficients of SDAR model, and the parameters of the RBFAR model is estimated by anoffline structured nonlinear parameter optimization method (SNPOM). After that, this model is applied to predict the data. Through the train and test for the real data of daily grid electricity of hydropower stationsin the left bank and right bank of Three Gorges as well as comparison with other classical algorithms, the forecasting accuracy and feasibility of RBFAR model are verified.