A Deep Learning-based Routing with Predicted Traffics in Space-Ground Integrated Heterogeneous Networks

With the evolution of integrated space-air-ground networks, Low Earth Orbit (LEO) satellite networks encounter scaling and dynamic challenges. This study employs GCN (Graph Convolutional Networks), LSTM (Long Short-Term Memory), and attention mechanisms to build a traffic forecasting model. It explores satellite networks' spatial characteristics using GCN to capture their topological and nodal dynamics. These characteristics merged with historical traffic data, are analyzed using LSTM for accurate temporal variation prediction. The attention mechanism enhances the model's focus, yielding more precise forecasts. Furthermore, this paper proposes a routing optimization algorithm based on traffic forecasting. This algorithm can identify the load distribution in the network in advance and dynamically adjust the data transmission paths based on predicted traffic data. This method effectively allocates resources, ensuring efficient and balanced data transmission under various load conditions.