Computation offloading strategy for UAV-assisted fog computing emergency communication system based on deep reinforcement learning

Abstract

With the continuous development of modern communication technologies, many computation-intensive and delay-sensitive applications have emerged. Unmanned Aerial Vehicle (UAV) can perform excellent and unique in emergency communication systems by virtue of its flexibility. In this paper, a UAV-assisted fog computing emergency communication system is designed, wherein a UAV that has computing capabilities can act as a UAV Fog Access Point (UAV-FAP) and offer surrounding mobile devices (MDs) computational offloading services. The objective is to minimize the maximum processing delay of the system. The optimization problem of jointly optimizing MD scheduling, distribution of resources and UAV flight direction is proposed under energy consumption constraints. Considering that the problem is a mixed integer nonlinear programming (MINLP) problem, a computational offloading algorithm based on Deep Deterministic Policy Gradient (DDPG) is proposed to solve the non-convex problem. And the Switches from Adam to SGD (SWATS) algorithm is also invoked to update the neural network parameters with the optimizer to avoid falling into local optima to some extent. Simulation results show that the DDPG-SWATS joint algorithm converges quickly and has a small processing delay compared with the benchmark algorithm.