Joint Service Caching and Computation Offloading Strategy Based on Deep Reinforcement Learning

With the development of vehicular edge computing (VEC), vehicles are unable to independently process all of the computational tasks they generate and must offload tasks to edge servers. However, in remote areas, the low deployment density of edge servers makes it difficult to meet the computational needs of vehicular users. By introducing satellite-assisted collaborative computing, not only can the issue of sparse edge server deployment be alleviated, but it also effectively reduces the computational load on Terrestrial networks, enhancing the overall system's computational capacity and service coverage. However, the time-varying characteristics of vehicular tasks and the limited cache space of edge servers necessitate the formulation of efficient service caching strategies. The close coupling between service caching and computation offloading decisions complicates the optimization strategy design. To address this, this paper constructs a satellite-Terrestrial collaborative vehicular network environment (CSTIOV). First, it considers the time-varying characteristics of vehicular task requests. Second, to fully utilize the resources of edge servers and minimize the long-term energy consumption of the system, this paper proposes a deep deterministic policy gradient-based algorithm (DDPG-CO), which improves system performance by jointly optimizing computation offloading, service caching, and resource allocation strategies. Simulation results show that the DDPG-CO algorithm significantly reduces system energy consumption and outperforms other benchmark algorithms.