Deep Reinforcement Learning for Computation Offloading and Resource Allocation in Satellite-Terrestrial Integrated Networks

Satellite mobile edge computing (SMEC) enhanced satellite-terrestrial integrated networks (STIN) have attracted intensive attention to obtain seamless coverage and provide on-demand computation services. However, the cooperative task execution among low earth orbit (LEO) satellites is largely ignored in the SMEC-STIN. In this paper, we explore a hybrid cloud and edge computing architecture of the SMEC-STIN with coordinated task processing among neighboring LEO satellites. We investigate the computation offloading and resource allocation strategies to minimize the long-term cost in terms of a trade-off between task execution latency and energy consumption. We formulate the optimization problem as a Markov decision process and design a proximal policy optimization based deep reinforcement learning method to approximate the optimal solution with robust training stability and low storage demand. Simulation results validate the effectiveness of our proposed method.