Mobile Edge Collaborative Caching Algorithm Based on Asynchronous Federation Popularity Prediction and Soft Multi-Agent Actor-Critic

Mobile edge collaborative caching is a crucial technology for optimizing the performance of Internet of Vehicles (IoVs) network services. However, its application in IoVs environments faces several challenges, including limited resources, privacy concerns, and the need for algorithms adaptable to heterogeneous environments. Addressing the limitations of vehicle caching and computing power in IoVs, which prevent comprehensive content caching, we propose a mobile edge collaborative caching algorithm leveraging asynchronous federated popularity prediction and a soft multi-agent actor-critic method (AFPP-SMAAC). Firstly, we introduce an asynchronous federated learning algorithm to enhance the accuracy of the global model. Secondly, we develop a popularity-based content prediction algorithm, utilizing the global model to proactively identify the most popular content for vehicle edge calculations and cache it in a strategically selected Road Side Unit (RSU). And finally, we design a soft multi-agent actor-critic algorithm to optimize collaborative caching decisions, aiming to minimize content transmission delay and maximize cache hit rates. Simulation results demonstrate that the AFPP-SMAAC algorithm outperforms existing baseline caching algorithms.