Task placement and resource allocation for UAV and edge computing supported transportation systems

Task placement and resource allocation for UAV and edge computing supported transportation systems

연구 분야: Networking

학회: The Journal of Supercomputing

Air-ground Integrated Networks (AGINs) supported by Mobile Edge Computing (MEC) have shown great potential in Intelligent Railway Systems (IRSs). In this paper, we investigate task placement, task replacement, and resource management issues in an UAV-supported IRS, utilizing Unmanned Aerial Vehicles (UAVs) as edge nodes to address tasks offloaded from ground Internet of Things (IoT) devices, thus to maximize the success rate of task execution. Considering the complexity and dynamics of transportation systems, we adopt a Multi-Agent Deep Deterministic Policy Gradients(MADDPG) based Deep Reinforcement Learning (DRL) algorithm to cope with the changing environmental conditions and evolving task requirements in the transportation system, and achieve maximization of objective function. Since MADDPG is designed for problems with only continuous variables, we tailor the output into discrete variables in training according to probabilities, and in the final decision making according to the rounding principle, and thus improve it for our problem with both discrete and continuous variables. Through simulations, we demonstrate that the improved MADDPG algorithm exhibits fast convergence characteristics, and also performs well in maximizing the task execution success rate of the system.