LLM-based task offloading and resource allocation in satellite edge computing networks

Satellite Mobile Edge Computing (MEC) networks offer a promising solution for delivering global services to terrestrial Internet of Things (IoT) terminals in 5 G and beyond. However, satellite MEC systems face challenges such as underutilization of resources and task congestion, leading to resource waste and increased latency. In this paper, we investigate the joint resource allocation and task offloading problem in multi-satellite MEC networks, aiming to minimize the average latency of IoT terminals. To solve the joint optimization problem involving IoT terminals' task offloading decisions, uplink transmission power and sub-channel allocation, and satellite computation resource allocation, we propose an iterative optimization algorithm that uses the Lagrange multipliers method to optimize the satellite computation resource allocation and a Large Language Model (LLM) based optimizer to optimize the other variables in each iteration. Prompts and templated parameters are designed to enhance the LLM's inference accuracy and generalization capability across scenarios with varying numbers of satellites and IoT terminals. Simulation results show that our proposed LLM-based algorithm outperforms benchmark algorithms in convergence speed and average latency of IoT terminals.