期刊
IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY
卷 69, 期 12, 页码 16067-16081出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/TVT.2020.3041929
关键词
Task analysis; Resource management; Servers; Vehicle dynamics; Dynamic scheduling; Optimization; Computational modeling; Vehicular fog computing (VFC); task offloading; task priority; dynamic pricing; soft actor-critic (SAC)
资金
- National Key Research and Development Program of China [2016YFB0100902]
- National Natural Science Foundation of China [62 071 268]
- China Postdoctoral Science Foundation [2020T130357]
Vehicular fog computing (VFC) has been expected as a promising scheme that can increase the computational capability of vehicles without relying on servers. Comparing with accessing the remote cloud, VFC is suitable for delay-sensitive tasks because of its low-latency vehicle-to-vehicle (V2V) transmission. However, due to the dynamic vehicular environment, how to motivate vehicles to share their idle computing resource while simultaneously evaluating the service availability of vehicles in terms of vehicle mobility and vehicular computational capability in heterogeneous vehicular networks is a main challenge. Meanwhile, tasks with different priorities of a vehicle should be processed with different efficiencies. In this work, we propose a task offloading scheme in the context of VFC, where vehicles are incentivized to share their idle computing resource by dynamic pricing, which comprehensively considers the mobility of vehicles, the task priority, and the service availability of vehicles. Given that the policy of task offloading depends on the state of the dynamic vehicular environment, we formulate the task offloading problem as a Markov decision process (MDP) aiming at maximizing the mean latency-aware utility of tasks in a period. To solve this problem, we develop a soft actor-critic (SAC) based deep reinforcement learning (DRL) algorithm for the sake of maximizing both the expected reward and the entropy of policy. Finally, extensive simulation results validate the effectiveness and superiority of our proposed scheme benchmarked with traditional algorithms.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据