Federated Reinforcement Learning-Driven Offloading Strategy for Edge Computing
Xiao Chen · 2024
In vehicular networks, Mobile Edge Computing (MEC) has become a crucial paradigm. Optimizing task offloading decisions is a key challenge in vehicular edge environments. Effective offloading strategies can significantly enhance system performance by minimizing latency and energy usage. To tackle the complexities of dynamic vehicular scenarios, we propose a novel approach that models the offloading optimization as a Markov Decision Process (MDP). Our methodology combines the strengths of reinforcement learning and federated learning paradigms. This integration results in an innovative algorithm: the Federated Learning-Twin Delayed Deep Deterministic Policy Gradient (FL-TD3). Experimental evaluations demonstrate that FL-TD3 achieves rapid convergence and superior training stability. This approach surpasses existing methods in minimizing both computation latency and energy use in vehicular edge environments.