In response to the problems of resource waste and long propagation time delay in traditional satellite power control methods, this paper proposes a satellite adaptive power control method based on Online Gate Recurrent Unit(Online-GRU) channel prediction, which solves the cumulative error of offline prediction algorithms by updating network parameters through online training. The simulation results show that the proposed online training algorithm improves the prediction accuracy by 38.30% compared to offline algorithms, saves 63.21% of training time compared to Online Long Short Term Memory(Online-LSTM), and saves 55.74% of transmission power compared to the fixed transmission power method. At the same time, the proposed adaptive power control method has better robustness compared to the adaptive power control method based on ground timing feedback channel state.