Affiliation of Author(s):合肥工业大学电气与自动化工程学院；安徽省工业自动化工程技术研究中心

Journal:IEEE Internet of Things Journal

Key Words:Deep deterministic policy gradient; resource allocation; task offloading; two-timescale; vehicular edge computing

Abstract:Vehicular edge-computing (VEC) server provides vehicles with sufficient computational resources to meet the low-latency requirements of their tasks. In this article, we study a novel partial offloading scheme in a VEC network, where the task can be unevenly partitioned into multiple subtasks. Different from existing work, we account for the nonarbitrary divisibility of tasks and enable parallel execution of task uploading and computing during offloading. Taking into account task generation time and small-scale channel gain variations at different time scales, a two-timescale dynamic offloading problem is studied with the objective of minimizing the average processing delay of all tasks by jointly optimizing resource allocation and the number of subtasks offloaded. The problem turns out to be a min-max optimization problem. To solve it, a deep deterministic policy gradient (DDPG)-based two-timescale task offloading and resource allocation (DDPG-TTORA) algorithm is proposed to obtain resource allocation decisions at each long time slot and subtask-offloading decisions at each short time slot. Simulation results demonstrated that the proposed DDPG-TTORA algorithm can reduce the average processing delay by 36.75% compared to the full partial offloading (FPO) algorithm.

Note:中科院二区TOP期刊

Indexed by:Journal paper

Translation or Not:no

Date of Publication:2026-04-20

Included Journals:SCI