Abstract: A new method of satellite orbit prediction based on artificial neural network (ANN) model is proposed in order to improve the orbital precision for BeiDou satellites. Several factors influencing compensation effect were analyzed particularly. Short-term, middle-term and long-term forecasting project were established based on adjusting compensable wave. The experiment results showed that, the orbit prediction errors based on ANN model were less than those based on dynamical model. In short-term orbit prediction, compensable wave needed to be moved when the time between training sample and current time were greater than 10 d, and compensable wave needed to be moved in middle-term and long-term orbit prediction. Comparison with that of un-adjusting compensable wave, rates of improvement were respectively improved 2.3%, 6.7%, 10% based on adjusting compensable wave when the prediction arcs are 8, 15, 30 d.