Dynamic User Association and Computation Offloading in Satellite Edge Computing Networks via Deep Reinforcement Learning
Hangyu Zhang, Hongbo Zhao, Rongke Liu, Xiangqiang Gao, Shenzhan Xu
Abstract
Satellite mobile edge computing (SMEC) deployed on ultra-dense low Earth orbit (LEO) satellites with high throughput and low latency can provide ubiquitous computing services closer to the user side. However, considering the highly dynamic and limited resources of LEO constellations, a joint strategy for accessing and offloading of ground users becomes difficult under overlapping satellite coverage. In this paper, a joint optimization method of dynamic user association and computation offloading for SMEC is proposed. Terrestrial users with random and diverse tasks adaptively access the optimal associated satellite under time-varying channel conditions, and offload to a satellite with sufficient remaining computing capability for load balancing in the SMEC network with inter-satellite cooperation. Furthermore, an evolutionary algorithm based on deep Q-network (DQN) is designed to jointly optimize the decisions of associated and offloading satellites and the allocation of computing resources, which enables energy-efficient strategies while meeting task latency and SMEC resource constraints. The method learns multi-dimensional actions intelligently and synchronously by improving network structure. The simulation results show that the proposed scheme can effectively reduce the system energy consumption by ensuring that the task is completed on demand, and outperform the benchmark algorithms.