Improving satellite network efficiency with terminal traffic prediction and SQP-SRA algorithm

To address the low resource utilization in satellite networks caused by heterogeneous regional traffic demands, this paper proposes a resource allocation strategy for LEO satellite internet based on terminal traffic prediction. An improved LSTM-GRU hybrid model is developed using real-world datasets to forecast ground traffic, accounting for periodic patterns and weather effects. A leaseable EOSN differentiated transmission framework is designed to enable targeted resource allocation and inter-satellite leasing, enhancing network coverage. To optimize data transmission ratios, user bandwidth, and service pricing, we introduce a sequential quadratic programming-based satellite resource allocation (SQP-SRA) algorithm that balances latency and energy consumption. Compared with LSTM, GRU, Transformer, and wavelet neural networks, the proposed model reduces traffic prediction error by approximately 26%. Simulation results demonstrate that, relative to the DDTOA, FCFS, and TOMRA algorithms, the proposed strategy improves user benefits by approximately 60% and enhances satellite service provider revenues by approximately 80%.