基于强化学习的灾区应急无人机网络服务公平性最大化方案

打开文本图片集

关键词：无人机通信控制；公平性指数；深度强化学习；混合动作空间中图分类号：TP183 文献标志码：A 文章编号：1001-3695（2025）08-027-2452-08doi：10.19734/j.issn.1001-3695.2024.12.0498

Reinforcement learning-based approach for maximizing service fairness in emergency drone networks in disaster areas

LiHuaicheng，Peng Jian†，HuangWen，ShenQunli，Liao Sirui （ 61oo65，China）

Abstract：Existing methodsfor unmanned aerial vehicle （UAV）-based emergency communication services in disaster areas optimizenetwork performanceunderglobalenvironmental information.However，thesemethodssuferfromlownetworking eficiencyandunbalancedresourcealocation，which hinderthe maintenancestablecommunication services indynamicdisaster environments.As aresult，some users may not receivetimelyrescue.This paper addresed the problem maximizing UAV communication quality.The method modeled the problem as a POMDP and designed a deep reinforcement learning-based approach tooptimize UAV path planning andresource alocation.The methodused network throughput astheservicequalitymetricandJain’sfairnessindexasthebalancingcriterion.Itdevelopedareward functionmechanismbasedonobjectivedecoupling，and constructeda parameterized deep graph reinforcement learning network toachieve joint optimization UAV trajectoryplanningandresource allcation.Extensivecomparativeexperiments wereconductedunder16dferent simulationconditions.The proposed method significantlyoutperformed four baselinemethods，improving the fairness index by 9.6% and demonstrating effectiveness across multiple performance metrics.

Key words：UAV communication control；fairness index；deep reinforcement learning；hybrid action space

0 引言

无人机通过搭载通信模块，作为空中基站为地面用户提供临时通信服务已经得到了广泛的应用，这对搭建灾区救援的应急网络是一项有前景的技术[1，2]。（剩余20325字）