Distributed Task Offloading Method Based on Federated Reinforcement Learning in Vehicular Networks with Incomplete Information

Rui Cao, Zhengchang Song, Bingxin Niu, Junhua Gu, Chunjie Li

Abstract: With the development of Internet of Vehicles, a large number of different types of applications have emerged. However, given the limited computing power of vehicles, many tasks cannot be completed within specified time. The task offloading method based on mobile edge computing (MEC) effectively solves the problem. However, the existing research does not fully consider user needs and has the disadvantages of limited application scope and slow training speed, so it is not suitable for high-speed vehicle scenarios. Considering the heterogeneity of tasks and the different needs of users, this paper first designs a dynamic weighting method for delay and vehicle energy consumption, and then proposes a distributed algorithm FDQN based on Deep Q-Network (DQN) and federated learning. The algorithm does not require any information about the base station and can utilize user collaboration to achieve fast convergence. The experimental results show that our algorithm can achieve best optimization effect under different vehicle energy and speeds.