Abstract : As the general mobile edge computing (MEC) scheme cannot adequately handle the emergency communication requirements in vehicular networks, unmanned aerial vehicle (UAV)-assisted vehicular edge computing networks (VECNs) are envisioned as the reliable and cost-efficient paradigm for the mobility and flexibility of UAVs. UAVs can perform as the temporary base stations to provide edge services for road vehicles with heavy traffic. However, it takes a long time and huge energy consumption for the UAV to fly from the stay charging station to the mission areas disorderly. In this paper, we design a predispatch UAV-assisted VECNs system to cope with the demand of vehicles in multiple traffic jams. We propose an optimal UAV flight trajectory algorithm based on the traffic situation awareness. The cloud computing center (CCC) server predicts the real-time traffic conditions, and assigns UAVs to different mission areas periodically. Then, a flight trajectory optimization problem is formulated to minimize the cost of UAVs, while both the UAV flying and turning energy costs are mainly considered. In addition, we propose a deep reinforcement learning(DRL)-based energy efficiency autonomous deployment strategy, to obtain the optimal hovering position of UAV at each assigned mission area. Simulation results demonstrate that our proposed method can obtain an optimal flight path and deployment of UAV with lower energy consumption.
Index terms : Deep reinforcement learning, energy efficiency, mobile edge computing, unmanned aerial vehicle relay