An Efficient Computation Offloading Strategy Based on Cloud-Edge Collaboration in Vehicular Edge Computing

Computation-intensive and latency-sensitive vehi-cle tasks continue to emerge with the repaid development of the Internet of Vehicles (IoV). Traditional cloud servers and single-point edge servers are unable to fulfill the demand for a large number of application services in a short period of time, resulting in the edge nodes having inadequate and im-balanced distribution of computing power in vehicular edge computing (VEC) networks. In response to the above difficul-ties, a cloud-edge collaboration hierarchical intelligent-driven VEC network architecture is first proposed, which utilizes the heterogeneous computing capabilities of cloud center, ag-gregation servers and MEC servers to achieve comprehensive collaboration and intelligent management of network re-sources. We then formulate the computation offloading strat-egy as an optimization problem that minimizes the total long-term cost of the system under communication and resource constraints, and transform the problem into a Markov decision process (MDP), taking into account the delay and energy consumption requirements of the computation tasks. Finally, considering the dynamic and stochastic nature of the VEC network, an efficient computation offloading strategy based on cloud-edge collaborative deep Q-network (CEC-DQN) is given to solve the MDP problem. Simulation results show that the proposed algorithm can significantly improve the VEC performance compared with the traditional single-point MEC offloading or random offloading algorithms.

I. Introduction

In recent years, VEC network have received considera-ble attention due to their wide range of applications. How-ever, the limited resources of onboard devices can not sup-port the increasing number of new applications, such as in-formation services, path navigation, and entertainment ap-plications [1]. In order to meet the diverse service require-ments of a large number of users in the VEC network, it is urgent to design efficient computation offloading strategies [2].