Abstract
The vehicular edge computing (VEC) can cache contents in different RSUs atthe network edge to support the real-time vehicular applications. In VEC, owingto the high-mobility characteristics of vehicles, it is necessary to cache theuser data in advance and learn the most popular and interesting contents forvehicular users. Since user data usually contains privacy information, usersare reluctant to share their data with others. To solve this problem,traditional federated learning (FL) needs to update the global modelsynchronously through aggregating all users' local models to protect users'privacy. However, vehicles may frequently drive out of the coverage area of theVEC before they achieve their local model trainings and thus the local modelscannot be uploaded as expected, which would reduce the accuracy of the globalmodel. In addition, the caching capacity of the local RSU is limited and thepopular contents are diverse, thus the size of the predicted popular contentsusually exceeds the cache capacity of the local RSU. Hence, the VEC shouldcache the predicted popular contents in different RSUs while considering thecontent transmission delay. In this paper, we consider the mobility of vehiclesand propose a cooperative Caching scheme in the VEC based on AsynchronousFederated and deep Reinforcement learning (CAFR). We first consider themobility of vehicles and propose an asynchronous FL algorithm to obtain anaccurate global model, and then propose an algorithm to predict the popularcontents based on the global model. In addition, we consider the mobility ofvehicles and propose a deep reinforcement learning algorithm to obtain theoptimal cooperative caching location for the predicted popular contents inorder to optimize the content transmission delay. Extensive experimentalresults have demonstrated that the CAFR scheme outperforms other baselinecaching schemes.