Asynchronous Federated Learning for Vehicular Edge Caching of Consumer Content

Edge caching is one of the key technologies, which enhances consumer content delivery and reduces service latency by caching consumer content and services on edge nodes. Learning-based caching algorithms have been proposed in the literature to achieve high caching efficiency. However, in the complicated task of edge caching for consumer content in vehicular networks, it is challenging to achieve a high cache hit rate and low consumer content delivery delay. This paper proposes AFRL, an Asynchronous Federated Learning with Deep Reinforcement Learning edge consumer content caching algorithm for vehicular networks. AFRL utilizes federated learning to collaboratively train a shared DRL agent among Roadside Units(RSUs), and an efficient asynchronous federated learning algorithm is also introduced to accelerate convergence and improve cache hit rates in dynamic environments. Simulation results demonstrate the superior performance of AFRL compared to traditional and state-of-the-art caching algorithms, showcasing its potential in handling varying traffic densities, achieving higher cache hit rates, and low consumer content delivery delay.