摘要详情

ID / 提交时间

57 / 2025-03-29 21:11:48

标题

Multi-Agent Reinforcement Learning Scheduling Strategy for UAV Swarms

关键字

Multi-Agent Reinforcement Learning,UAV Swarm Scheduling,Dynamic Task Allocation,Hybrid Reward Mechanism,Power Infrastructure Inspection

主题及专题

A11机器人

状态

全文待审

作者

Wang Renshu / 国网福建电力科学研究院

Wu Wenbin / 国网福建电力科学研究院

Zhang Weihao / 国网福建电力科学研究院

Chen Zhuolei / 国网福建电力科学研究院

Liang Manshu / 国网福建电力科学研究院

摘要

Unmanned aerial vehicle (UAV) swarms offer significant efficiency gains for power infrastructure inspection but face challenges in dynamic task allocation and real-time scheduling. This study proposes a hierarchical multi-agent reinforcement learning (MARL) framework integrating graph neural networks and a hybrid reward mechanism. Key innovations include global-local decision balance, efficient state encoding via graph topology, and a reward function combining short-term inspection efficiency with long-term maintenance forecasting. Experiments demonstrate a 35% improvement in inspection coverage and a 28% reduction in emergency response times compared to traditional methods.