摘要
针对多无人机协同侦察规划问题,由于自顶而下的研究方法有效降低了问题的求解难度,该方法逐渐成为主流研究方向。然而,当面对当前环境发生变化时,上述方法需要对问题进行重新优化求解,以至于实时性表现较差。为解决上述问题,文中提出基于聚类和强化学习的无人机群协同侦察任务规划,该方法将无人机荷载的探测半径考虑到侦察任务的聚类算法中,将探测半径作为范围限定,对整个任务区域进行划分,重新聚为K个子区域,并将子区域内的目标归类到一个中心目标,从而使得原始目标的聚类结果更鲁棒的同时有效降低任务的量级。此外,将影响协同任务的关键因素,例如无人机航行状态、存活几率以及环境变化等要素作为任务目标的约束项,构建协同侦察任务规划的优化模型;最后将奖励函数应用到协同任务求解中通过最大化奖励优化模型的性能,从而达到对环境等要素的良好适应性。
The top-down method has been a top trending way of multi-UAVs cooperative reconnaissance mission planning,because of effectively reducing the difficulty of solving this research.However,the existing top-down methods have to re-solve and re-optimize problem with changing current environmental condition,resulting in poor real-time performance.A multi-UAVs reconnaissance mission planning based on cluster and reinforcement learning method is proposed to slove the problem.Firstly,clustering method is optimized by using detection radius of UAV’s load to compute the number of clusters and to cluster tar-gets,which makes the clustering result of the target set stable and reduces the scale of mission number for following cooperative mission planning.Considering that environmental information,survival probability and drone endurance constraints are the key factors affecting cooperative task,they are as constraint con-ditionas to establish optimazation model.Finally,Combined this model with the reward function of rein-forcement learning,the performance can be optimized,and in this way,the model can achieve good adaptability under the complicated environment.
作者
刘敬蜀
吴嘉琪
刘旭波
LIU Jing⁃shu;WU Jia⁃qi;LIU Xu⁃bo(No.3 Laboratory,The Unit 91977 of PLA,Beijing 100036,China)
出处
《中国电子科学研究院学报》
北大核心
2023年第1期21-25,55,共6页
Journal of China Academy of Electronics and Information Technology
关键词
无人机群
协同侦察
任务规划
聚类
强化学习
drone swarm
cooperative reconnaissance
mission planning
cluster
reinforcement learning
