Deep Reinforcement Learning for Edge Caching and Content Delivery in Internet of Vehicles

To enable the emerging vehicular applications and multimedia services in an Internet of Vehicles (IoV) framework, edge caching is a promising paradigm which can cache content in proximity to vehicles, and thus alleviate heavy load on backhaul links and contribute in reducing transmission latency. However, in a multi-access vehicular network, complex content delivery and high mobility of vehicles introduce new challenges to support edge caching in a dynamic environment. Deep Reinforcement Learning (DRL) is an emerging technique to solve the issue with time-varying feature. In this paper, we utilize DRL to design an optimal vehicular edge caching and content delivery strategy for minimizing content delivery latency. We first propose a multiaccess edge caching and content delivery framework in vehicular networks. Then, we formulate the vehicular edge caching and content delivery problem and propose a novel DRL algorithm to solve it. Numerical results demonstrate the effectiveness of proposed DRL-based algorithm, compared to two benchmark solutions.