Dueling Double Deep Q-Network Based Computation Offloading and Resource Allocation Scheme for Internet of Vehicles

This paper investigates a computation offloading and resource allocation policy for multiple vehicle user equipments (VUEs) in the Internet of Vehicles (IoV). Aiming at balancing the delay and energy consumption during the offloading procedure, a Support Vector Machine (SVM) is initially adopted to classify the offloading tasks into two categories according to different delay and energy consumption requirements. Consequently, VUEs can choose to offload the tasks to the mobile edge computing (MEC) server or other VUEs for completion. In particular, to further decrease the task offloading time in the MEC processing mode, the non-orthogonal multiple access (NOMA) scheme is adopted, which makes it possible for the MEC server to serve two VUEs simultaneously on the same sub-channel. To minimize the total cost, a Dueling Double Deep Q-Network (D3QN) based resource allocation algorithm is proposed, which can allocate the corresponding radio or computing resources under different task processing modes. Simulation results demonstrate that the proposed scheme can effectively reduce the total offloading cost within the maximum delay tolerance compared with existing methods.