Satellite Network State Prediction and Route Planning Policy Using the MPNN Algorithm.

Network status prediction and efficient routing play important roles in low earth orbit (LEO) satellite networks. It is necessary to establish an accurate network model for state prediction in LEO satellite networks with high mobility. According to the predicted state, route planning is needed to satisfy requirements. In this work, we propose a satellite network state prediction and route planning method combining the Message Passing Neutral Network (MPNN) algorithm and Q-Leaning. Firstly, a network topology establishment method is proposed based on the idea of Virtual Node (VN). Then, the satellite network status is accurately predicted based on the MPNN algorithm with the attention mechanism. Q-Learning is further applied to obtain the optimal routing path after prediction. Finally, through simulation experiments, the accuracy and generalization of our algorithm is verified, and the route policy can be obtained.