首页 >

面向无人机集群协同的感知-通信-控制联合优化方法-自主智能无人系统专刊

许驰,李安顺,于海斌   

  1. 中国科学院沈阳自动化研究所
  • 收稿日期:2026-06-01 修回日期:2026-10-12 出版日期:2026-10-19 发布日期:2026-10-19
  • 通讯作者: 许驰
  • 基金资助:
    国家自然科学基金青年基金B类;辽宁省自然科学基金A类;辽宁省“兴辽英才”青年拔尖人才项目

Joint Optimization of Sensing, Communication, and Control for UAV Swarms Collaboration

  • Received:2026-06-01 Revised:2026-10-12 Online:2026-10-19 Published:2026-10-19
  • Supported by:
    工业无线控制网络技术与系统;工业互联网“通感算控智”协同理论与关键技术;工业智联网理论与技术

摘要: 针对无人机集群空地协同过程中的高实时、低功耗调度问题,构建了感知-通信-控制(S2C)联合优化模型,提出了一种融合图卷积网络(GCN)与近端策略优化(PPO)的优化算法。首先,充分考虑无人机集群动态拓扑频繁切换、多约束强耦合等特点,以同时最小化时延和能耗为目标,构建了关于感知频次、通信功率、飞行速度和航向角的联合优化问题,并将其建模为马尔可夫决策过程。其次,利用GCN对时变拓扑与链路特征进行编码,设计基于阈值感知的动态拓扑更新机制,并通过量化拓扑变化幅度触发局部消息传递更新,实现对无人机链路频繁通断与节点动态入网的高效自适应。在此基础上,基于PPO进行集中式多目标联合优化,并设计惩罚函数引导算法优先满足跟踪精度要求,动态权衡时延与能耗。仿真结果表明,所提算法较基线方法降低时延45.5%、降低能耗3.2%,轨迹跟踪误差稳定在30 m以内。

关键词: 无人机集群, 空地协同, 轨迹跟踪, 资源调度, 图卷积网络, 近端策略优化

Abstract: To satisfy the high-realtime and low-power scheduling requirements for air-ground collaboration of UAV swarm, this paper establishes a joint optimization model for Sensing-Communication-Control (S2C) and proposes an optimization algorithm that combines a Graph Convolutional Network (GCN) with Proximal Policy Optimization (PPO). Fully considering the characteristics of frequent dynamic topology changes and strongly coupled constraints, and aiming at minimizing latency and energy consumption, we first formulate a joint optimization problem with respect to perception frequency, transmit power, flighting speed, and heading angle, and model the problem Markov decision process. Then, GCN is adopted to encode time-varying topology and link features, and a threshold-aware dynamic topology update mechanism is designed. By quantifying topology variations and triggering local message-passing update, we realize high-efficiency adaption to frequent link interruptions and dynamic node access of UAVs. On this basis, PPO is employed for centralized multi-objective collaborative optimization. A penalty-based reward function is designed to prioritize tracking accuracy and dynamically balance latency and energy consumption. Simulation results show that, compared with baseline algorithms, the proposed algorithm reduces latency by 45.5% and energy consumption by 3.2%, and maintains the trajectory tracking error steadily within 30 m.

Key words: UAV swarm, air-ground collaboration, trajectory tracking, resource scheduling, graph convolutional network, proximal policy optimization