无线电工程2026,Vol.56Issue(7):1216-1224,9.DOI:10.3969/j.issn.1003-3106.2026.07.009
不完美先验下应急通信无人机贝叶斯在线资源优化
Bayesian Online Resource Optimization for Emergency Communication UAV with Imperfect Prior
邱荣兴 1李志坚 2石尧 2马昊 2魏康2
作者信息
- 1. 国防科技大学 信息通信学院,湖南 长沙 410073
- 2. 中国人民解放军31631 部队,广东 惠州 516008
- 折叠
摘要
Abstract
Unmanned Aerial Vehicle(UAV)serving as aerial base stations have emerged as a promising solution for restoring communication in post-disaster scenarios.In earthquake-affected areas,a UAV must serve heterogeneous users—rescue personnel with strict rate requirements and affected civilians with best-effort service—while operating under limited energy and relying only on a pre-disaster crowd heatmap as prior spatial information.The heatmap is inherently imperfect,as the earthquake reshapes population distribution in unknown ways.A Bayesian online learning framework,termed Bayesian Online Learning for UAV Emergency Communication(BOL-UEC),treats the pre-disaster heatmap as a prior distribution and progressively refines it through real-time user location observations.The framework decouples decision-making into two timescales:a slow timescale for Bayesian belief update and UAV trajectory planning,and a fast timescale for Orthogonal Frequency Division Multiple Access(OFDMA)subcarrier and power allocation with High Priority(HP)minimum rate guarantees and Jain's fairness constraints.Simulation results demonstrate that in the early mission phase,the cumulative rate of Low Priority(LP)users under BOL-UEC is improved by 13.6%compared with the prior-free baseline scheme adopting the ∈-decreasing greedy exploration strategy.The overall gain is 10.3%,with the gain narrowing as Bayesian online learning progressively overcomes the uninformative prior.BOL-UEC consistently maintains lower posterior Kullback-Leibler(KL)divergence and marginally higher Jain's fairness compared to all prior-free baselines.Comparisons against multiple exploration strategies decreasing ∈-greedy,Thompson Sampling(TS)and a simple follow-observations heuristic confirm that the heatmap prior's value is real and most consequential during the critical early deployment phase.关键词
无人机辅助通信/贝叶斯在线学习/资源分配/轨迹优化/人群热力图/用户优先级Key words
UAV-assisted communication/Bayesian online learning/resource allocation/trajectory optimization/crowd heatmap/user priority分类
信息技术与安全科学引用本文复制引用
邱荣兴,李志坚,石尧,马昊,魏康..不完美先验下应急通信无人机贝叶斯在线资源优化[J].无线电工程,2026,56(7):1216-1224,9.