LLM-Enhanced MARL for Multi-Hop UAV Networking: An Edge-Cloud Protocol Aware Framework
Yuhao Wu, Xiulin Qiu, Ping Wang, Hui Ye, Donghui Zhang, Jian Cui
Jiangsu University of Science and Technology York University
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
With the development of the integration of aerial, terrestrial, and satellite networks, Uncrewed Aerial Vehicles (UAVs) serve as a new type of aerial communication platform, playing a key role in Internet of Things (IoT) scenarios. However, multi-hop UAV networks still suffer from energy limits, bandwidth scarcity, dynamic topology, and unstable links. To address these challenges, this paper proposes an edge-cloud communication protocol framework (EC-LLM-MARL) based on Large Language Models (LLMs) and MARL. In this framework, the cloud side constructs a chain-of-thought to guide LLM reasoning, and distilled expert strategies are transferred to UAVs through knowledge distillation and fine-tuning, improving training efficiency and reducing communication overhead. Additionally, a Graph Attention Network (GAT) is employed to extract deep features of observation targets, while Multi-Agent Deep Deterministic Policy Gradient (MADDPG) is combined with a self-attention mechanism at the UAV end to enhance credit assignment and convergence efficiency in dynamic environments. Furthermore, the framework designs a differentiated reward mechanism for task load balancing among multiple UAVs, improving the scalability of the algorithm in large-scale network scenarios. Extensive simulations show that EC-LLM-MARL significantly accelerates training convergence and substantially outperforms existing methods, achieving a 77% increase in throughput, a 13% rise in coverage, and the highest remaining energy levels.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
工程UAV Applications and Optimization
IoT and Edge/Fog Computing · Mobile Ad Hoc Networks