On a Novel Content Edge Caching Approach based on Multi-Agent Federated Reinforcement Learning in Internet of Vehicles

Abstract

Driven by the emerging requirements of Internet of Vehicles (IoV), future vehicles are expected to have the ability to provide not only the autonomous driving services, but also the multimedia services for working and entertainment. The edge caching service enabled by the Road Side Units (RSUs) can complement the limited environment perceiving and computing ability of future vehicles to gather, pre-process, and cache the contents of driving assistance, work, and entertainments. In this paper, we use federated learning to learn the popularity variation tendency considering user preference in different districts and their concerns for privacy-preserving. We further split the possible contents into blocks and use completely cooperative multi-agent reinforcement learning based on Deep Q network to make a more flexible and accurate caching decision considering the various emergency levels and delay requirements of different contents. Numerical results demonstrate that the proposed method outperforms traditional caching strategies.