空天辅助车载边缘计算中基于深度强化学习的计算卸载策略OA

To address the computation offloading problem in aerospace-assisted vehicular edge computing scenarios,a deep reinforcement learning-based computation offloading strategy is proposed to support the computation offloading of the Internet of Vehicles in remote areas,considering the het-erogeneity of resources in the space-air-ground integrated network SAGIN)and the task dependency structure in intelligent vehicular applications.The strategy first constructs a network scenario for aero-space-assisted vehicular computing based on the characteristics of different devices,such as on-board terminals,unmanned aerial vehicles,and low Earth orbit satellites.Then,this computation offloading problem is formulated as a Markov decision process MDP).Finally,with the objective of minimizing the average delay and energy consumption during the offloading process,a computation offloading strategy is designed based on an improved deep deterministic policy gradient DDPG)algorithm.Ex-perimental results show that compared with the conventional reinforcement learning algorithms DDPG and deep Q-network DQN,the average application delay of the proposed strategy is reduced by 30.29％and 54.11％,while the average energy consumption is reduced by 38.76％and 57.12％respectively.