Energy-Efficient Task Offloading for High-Speed Trains with Multi-Orbit Satellite Networks

Abstract

Given the constrained computational capabilities of onboard processors and train-mounted sensing devices, which limit the efficient execution of computation-intensive tasks, this study proposes a satellite-assisted computation offloading framework to enhance processing efficiency in railway networks. We first analyze the energy consumption and latency characteristics of train task processing under various satellite switching scenarios and green energy harvesting conditions. To this end, a multi-orbit satellite system model is developed, incorporating orbit-specific handover dynamics and energy harvesting features. Based on this model, we design a flexible and adaptive computation offloading framework and introduce a Hierarchical Reinforcement Learning (HRL) algorithm to optimize the offloading strategy in dynamic satellite network environments. Simulation results demonstrate the effective convergence of the HRL algorithm, showing its ability to progressively refine the offloading strategy and achieve stable, energy-efficient operation. Moreover, the proposed framework significantly reduces overall energy consumption while improving task execution reliability across diverse deployment scenarios.