Research on Task Offloading Based on Deep Reinforcement Learning for Internet of Vehicles

Abstract

Mobile Edge Computing (MEC) is a promising technology that facilitates the computational offloading and resource allocation in the Internet of Vehicles (IoV) environment. When the mobile device is not capable enough to meet its own demands for data processing, the task will be offloaded to the MEC server, which can effectively relieve the network pressure, meet the multi-task computing requirements, and ensure the quality of service (QoS). Via multi-user and multi-MEC servers, this paper proposes the Q-Learning task offloading strategy based on the improved deep reinforcement learning policy(IDRLP) to obtain an optimal strategy for task offloading and resource allocation. Simulation results suggest that the proposed algorithm compared with other benchmark schemes has better performance in terms of delay, energy consumption and system weighted cost, even with different tasks, users and data sizes.