引用本文:张涛,刘威,王锐,李凯文,徐万里.考虑无线充电的无人机路径在线规划[J].控制理论与应用,2024,41(1):30~38.[点击复制]
ZHANG Tao,LIU Wei,WANG Rui,LI Kai-wen,XU Wan-li.Online path planning for unmanned aerial vehicles considering wireless charging[J].Control Theory and Technology,2024,41(1):30~38.[点击复制]
考虑无线充电的无人机路径在线规划
Online path planning for unmanned aerial vehicles considering wireless charging
摘要点击 2009  全文点击 2103  投稿时间：2022-04-29  修订日期：2022-11-01
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2023.20330
  2024,41(1):30-38
中文关键词  深度强化学习  无人机  智能优化  无线充电
英文关键词  deep reinforcement learning  unmanned aerial vehicles  intelligent optimization  wireless charging
基金项目  国家自然科学基金项目(72071205)资助.
作者单位邮编
张涛 国防科技大学 410073
刘威 国防科技大学 
王锐 国防科技大学 
李凯文* 国防科技大学 410073
徐万里 军事科学院 
中文摘要
      近年来, 无人机在物流、通信、军事任务、灾害救援等领域中展现出了巨大的应用潜力, 然而无人机的续航 能力是制约其使用的重大因素, 在无线充电技术不断突破和发展的背景下, 本文基于深度强化学习方法, 提出了一 种考虑无线充电的无人机路径在线优化方法, 通过无线充电技术提高无人机的任务能力. 首先, 对无人机功耗模型 和无线充电模型进行了构建, 根据无人机的荷电状态约束, 设计了一种基于动态上下文向量的深度神经网络模型, 通过编码器和解码器的模型架构, 实现无人机路径的直接构造, 通过深度强化学习方法对模型进行离线训练, 从而 应用于考虑无线充电的无人机任务路径在线优化. 文本通过与传统优化方法和深度强化学习方法进行实验对比, 所提方法在CPU算力和GPU算力下分别实现了4倍以及100倍以上求解速度的提升.
英文摘要
      Recently, unmanned aerial vehicles (UAVs) have shown great potentials in the fields of logistics, communication, military mission, disaster rescue, etc. However, the poor endurance of UAVs is a major problem that restricts their use. With the development of wireless charging, this paper proposes an online UAV path planning method considering wireless charging based on the deep reinforcement learning. The mission capability of UAVs can be improved by applying wireless charging. We first construct the UAV power consumption model and the wireless charging model. A deep neural network model with dynamic context is designed according to the power constraints of the UAV. The UAV path can be constructed by the encoder-decoder architecture of the model. The model is trained offline through deep reinforcement learning, and is applied to the online optimization of the UAV path. Experimental results show that, the solving speed of the proposed method is more than four times and a hundred times faster than traditional optimization and deep reinforcement learning methods on CPU and GPU, respectively.