UAV-Assisted Fog Computing Offloading Strategy Based on Deep Reinforcement Learning

With the arrival of the Internet of Things era, unmanned aerial vehicle (UAV) communication become more popular due to their high maneuverability and low operating costs. In this paper, a UAV equipped a fog computing server used to provide computing offloading services for user equipments (UEs). Under the constraint of certain energy consumption, we jointly optimize user scheduling, computing resource allocation, UAV flight direction and speed to achieve the optimization goal of minimizing the maximum processing delay. The deep deterministic policy gradient (DDPG) offloading algorithm improved by a priority experience replay (PER) pool can increase the probability of high-value sampling and solve the problem of sparse rewards. We introduce the DDPG-PER offloading algorithm into this scene to achieve minimize the maximum processing delay and design experiments to compare with other algorithms. Experiments show that the DDPG-PER offloading algorithm can quickly achieve the optimization goals of convergence. The proposed DDPG-PER algorithm can faster achieve convergence always has lower delay compared with other benchmark algorithms in different task size. Meanwhile, in different computing capability of UAV, it has a better performance on offloading ratio and decreasing delay compared with the other benchmark algorithm.