Dynamic Resource Scheduling for UAV Systems in Multi-Scenario Tasks Based on MADDPG Algorithm

Abstract

Unmanned Aerial Vehicle (UAV) systems face critical resource scheduling challenges in multi-scenario tasks such as disaster rescue, logistics delivery, and facility inspection. Traditional static scheduling methods fail to accommodate dynamic resource demands and real-time priority adjustments. This paper proposes a dynamic scheduling model based on Multi-Agent Deep Deterministic Policy Gradient (MADDPG) for four types of heterogeneous support resources: aircraft external configurations, ground control stations, ground support equipment, and associated tools. The model integrates a task priority assessment mechanism and a real-time resource state perception module, employing centralized training with decentralized execution (CTDE) architecture. Rather than proposing a fundamentally new learning algorithm, we engineer a problem-tailored integration of difference rewards, prioritized experience replay, and adaptive exploration that enables autonomous agents to perform effective multi-resource collaborative scheduling under decentralized execution. Simulation results show that the proposed model attains a task completion rate of 95.7%, resource utilization of 91.5%, and average response time of 4.2min, consistently outperforming representative baselines including MAPPO, QMIX, and Attention-MADDPG under identical simulation settings.