Deep Deterministic Policy Gradient Algorithm for Space/Aerial-Assisted Computation Offloading

Space-air-ground integrated network (SAGIN) has been envisioned as a promising architecture and computation offloading is a challenging issue, with the growing demand for computation-intensive applications in remote area. In this paper, we investigate a SAGIN edge computing architecture considering the energy consumption and delay of computation offloading, in which ground users can determine whether take advantage of edge server mounted on the unmanned aerial vehicle and satellite for partial offloading or not. Specifically, the optimization problem of minimizing the total cost is formulated as a Markov decision process, and we proposed a deep reinforcement learning-based method to derive the near-optimal policy, adopting the deep deterministic policy gradient (DDPG) algorithm to handle the large state space and continuous action space. Finally, simulation results demonstrate that the partial offloading scheme learned from proposed algorithm can substantially reduce the user devices’ total cost as compared to other greedy policies, and its performance is better than the binary offloading scheme learned from Deep Q-learning algorithm.