Computation offloading and tasks scheduling for the internet of vehicles in edge computing: A deep reinforcement learning-based pointer network approach

Abstract

In the Internet of Vehicles, vehicles can offload computation tasks to edge servers for execution. So, execution delay of tasks and energy consumption of vehicles can be reduced. However, the current research on computation offloading has not fully considered the priority of computation offloading and tasks scheduling within the server. In this regard, a computation offloading and task scheduling scheme based on pointer network was proposed in the paper. This scheme can maximize the number of offloading executions of computation tasks with limited computing resources of the edge server while satisfying the priority of computation offloading. First, depending on whether the task can be executed in the vehicle's device, we divided the task into two types and gave a two-stage offloading policy. Next, according to the characteristics of the scheduling problem, a task offloading decision and scheduling scheme based on pointer network was proposed. Then, considering the uncertainty of the number of tasks, the extra time delay caused by task offloading, the waiting time of tasks, and the complexity of task scheduling and computing resource allocation, a deep reinforcement learning algorithm was used to train the pointer network. Finally, the trained pointer network was used for task offloading decision-making and scheduling. The experimental results revealed the effectiveness of the scheme proposed in this study.