Research on Satellite Network Traffic Prediction Algorithm Based on Gray Wolf Algorithm Optimizing GRU and Spatiotemporal Analysis

Accurate traffic prediction is crucial for the performance improvement of satellite networks, which can help balance network load, optimize routing, and effectively improve network performance. To better plan traffic while capturing the spatial and temporal dependencies of traffic, we propose a new neural network-based traffic prediction method for satellite networks, namely the temporal graph convolutional attention (ACW-GRU) model, which is combined with the Graph Convolutional Neural Network (GCN), the Gated Recurrent Unit (GRU), and the Soft Attention mechanism. The GCN is used to analyze the complex topology to capture spatial correlation, and the GRU is used to analyze the dynamic changes of traffic data to capture temporal correlation, while the Grey Wolf Optimization (GWO) is used to optimize the hyperparameters of the GRU model, and the soft attention mechanism is used to integrate the global temporal information and adjust the importance at different time points. The importance of the traffic data at distant time points is also not ignored, thus improving the accuracy of the prediction. The simulation results show that the ACW-GRU model is superior in fitting the flow data compared with the traditional GRU, LSTM, and FARIMA models, and its root mean square error (RMSE) is reduced by 24.7%, 40.4%, and 51.2%, respectively. Therefore, the ACW-GRU model can predict the traffic data of the satellite network more accurately, thus providing clearer guidance for defining the demand of applications and the allocation of network resources.