| 基于MDP‑MAPPO驱动的多无人机路由实现可持续能源与延迟优化 |
| Sustainable energy and latency optimization of multi‑UAV routing based on MDP‑MAPPO |
| |
| DOI: |
| 中文关键词: 强化学习;车载网络;边缘计算;路径规划;数字孪生 |
| 英文关键词:reinforcement learning; vehicular networks; edge computing; path planning; digital twins |
| 基金项目:国网浙江电力移动化电力数字空间地图支撑能力优化提升实施项目(SGIT0000YXXX2200949)资助项目 |
| 作者 | 单位 | | 林黎鸣 | 1. 国网思极位置服务有限公司,北京 102211
2. 东南大学 仪器科学与工程学院,江苏 南京 210096 | | 陈思光 | 河海大学 计算机与软件学院,江苏 南京 210024 |
|
| 摘要点击次数: 196 |
| 全文下载次数: 48 |
| 中文摘要: |
| 随着车载网络对高质量服务和低延迟的需求不断攀升,引入无人机可解决计算与资源失衡
导致的延迟、拥塞等问题,但现行轨迹规划与任务卸载方案在降低能耗、提高实时性上存在不足,
难以适配车载网络数据的动态特性。文中提出 PPO算法,通过优化无人机飞行轨迹、设备关联及
任务卸载比例缓解系统延迟和能耗挑战,构建基于数字孪生的车载网络边缘计算架构,依托
MAPPO算法设计多无人机路径规划算法(MDP-MAPPO),在动态场景下融合数字孪生与强化学习
优化无人机轨迹。实验表明,该算法在收敛性上优于粒子群优化和A-star算法,且延迟更低、能耗
更小。 |
| 英文摘要: |
| The demand for high-quality services and low latency in vehicular networks is rapidly increasing. Integrating unmanned aerial vehicles (UAVs) helps resolve latency and congestion caused by computational/resource imbalances. Yet existing trajectory planning and task offloading schemes lack effectiveness in cutting energy use and boosting real-time responsiveness, struggling to adapt to vehicular networks’ dynamic data. To tackle these challenges, we propose a proximal policy optimization (PPO)-based approach, mitigating latency and energy problems by optimizing UAV trajectories, device associations, and task offloading ratios. First, we build a digital twin-enabled edge computing architecture for
such networks. Second, we design a multi-UAV path planning algorithm, named MDP-MAPPO, based
on the multi-agent PPO (MAPPO) framework. In dynamic scenarios, MDP-MAPPO optimizes UAV trajectories via integrating digital twin technology and reinforcement learning. Experimental results show
that MDP-MAPPO outperforms particle swarm optimization (PSO) and A-star in convergence speed,
with lower latency and less energy cost. |
| 查看全文 查看/发表评论 附件 |