Dynamic Task Offloading and Resource Allocation for Vehicular Edge Computing Networks Based on Deep Reinforcement Learning

Vehicular edge computing (VEC) server allocates computing resource blocks (CRBs) with different CPU frequencies to process tasks with different low latency requirements. During task processing, some task vehicles (TaVs) and CRBs may finish processing tasks in advance, and their computing resources are idle. However, the existing work has not taken into account reusing these idle computing resources to handle other ongoing tasks, causing low computing resource utilization efficiency. To address this, considering the dynamic nature of the VEC networks, a task re-scheduling problem is formulated to minimize the task completion delay to realize dynamic task offloading and resource allocation (DTORA). To solve the problem, we divide it into a joint task offloading and computing resource allocation problem and a joint subtask re-scheduling and power allocation problem. Dealing with the first problem, a joint task offloading and computing resource allocation (JTOCA) algorithm is proposed. Given the computing resource allocation decision, a deep deterministic policy gradient (DDPG)-based joint subtask re-scheduling and power allocation (DDPG-JSRPA) algorithm is proposed to solve the second problem. Simulation results demonstrated that the proposed DTORA algorithm can reduce the system delay by 28% compared to the full partial offloading (FPO) algorithm.