Deep Reinforcement Learning Based Traffic Offloading Scheme for Vehicular Networks

With the emergence of pervasive mobile devices, mobile cloud computing cannot fully meet the user demands, which promotes the birth of Mobile Edge Computing (MEC). Tasks could be offloaded to the MEC servers when the ability of mobile devices to process data does not satisfy its own needs. With the introduction of 5G and the development of Internet of Vehicles (IoV), the data generated by vehicles and passengers would require more computing tasks. In this paper, we use the deep reinforcement learning based method to offload the computation tasks by MEC. The evaluation of single user mobile edge offloading is first implemented, and then two deep reinforcement learning based algorithms are compared and analyzed. Then the comparison experiments are extended to the multi-user situation. After that, the suitable learning rates of computation offloading for IoV in MEC using deep deterministic policy gradient algorithm can be found. The experimental results demonstrate the efficiency of the designed offloading scheme.