Asynchronous Federated Based Vehicular Edge Computation Offloading

The expansion of vehicle-to-everything(V2X) systems has grown substantially in recent years due to technological breakthroughs like vehicle edge computing (VEC) and 5G. The rapidly developing field of VEC transfers computationally demanding activities to a nearby VEC server, allowing real-time applications for vehicles. Conventional deep reinforcement learning (DRL) algorithms are inadequate for addressing privacy concerns when outsourcing sensitive data activities. This work introduces an asynchronous federated deep reinforcement learning (AFDRL) approach for task offloading techniques to maximize computation rate while ensuring queue stability and data privacy. To tackle the issue, our research examines computation and queue models for executing tasks at the vehicle or roadside unit (RSU). We introduce a Lyapunov-enhanced deep reinforcement learning approach to address the optimization issue. The outcomes of the simulation show that our proposed approach may significantly improve the computation rate and maintain queue stability in the context of task outsourcing issues, as compared to several baseline methods.