极地研究

极地研究 ›› 2023, Vol. 35 ›› Issue (3): 392-404.DOI: 10.13679/j.jdyj.20220428

• 研究论文 • 上一篇    下一篇

基于联邦学习的中山站移动传感单元启发式路径规划算法研究

王煜尘1,2  祝标2  郭井学2  窦银科1  姚旭2  孙阳2   

  1. 1太原理工大学电气与动力工程学院, 山西 太原 030024;
    2中国极地研究中心(中国极地研究所), 上海 200136
  • 出版日期:2023-09-30 发布日期:2023-09-30
  • 通讯作者: 窦银科，孙阳
  • 作者简介:王煜尘, 男, 1995年生。博士, 主要从事极端环境传感器及机器学习算法方向研究。E-mail: tyut_yuchenwang@163.com
  • 基金资助:

    南极中山雪冰和空间特殊环境与灾害国家野外科学观测研究站(121163000000190015)资助

Application of a heuristic path planning algorithm for mobile sensing units in Zhongshan Station based on a federated learning mechanism

Wang Yuchen1,2, Zhu Biao2, Guo Jingxue2, Dou Yinke1, Yao Xu2, Sun Yang2   

  1. 1College of Electrical and Power Engineering, Taiyuan University of Technology, Taiyuan 030024, China; 
    2Polar Research Institute of China, Shanghai 200136, China
  • Online:2023-09-30 Published:2023-09-30

PDF (PC)

539

摘要：

启发式算法被广泛用于移动巡检单元的路径规划中。然而在一些特殊场合(例如南极中山站), 有限的通信带宽、能源和计算能力要求移动巡检单元在路径规划算法中更有效率。为了解决上述困难, 本研究提出了1个网络交换和分布式通信设施的设计方案, 并将其作为实现数字孪生传感网络的基础。同时, 本研究提出了1种改进型灰狼优化的路径规划算法, 结合联邦学习机制, 以提高路径规划算法的效率, 减少资源消耗。通过一系列的仿真研究和南极中山站现场试验, 验证了该算法在启发式全局路径规划、规划成本评估和区域动态路径规划方面取得了良好的性能。本研究设计的硬件平台功能符合实际任务要求, 新型启发式路径规划算法优于同类其他算法, 联邦学习机制提高了规划算法中参数设置的效率, 算法模型使南极中山站移动巡检单元的路径规划更加高效和可靠。

关键词:

路径规划, 启发式算法, 南极中山站, 联邦学习

Abstract:

Heuristic algorithms are widely used in path planning for mobile units. However, in specific situations (e.g., Zhongshan Station in Antarctic), restrictions in communication bandwidth, available energy, and computing power require more efficiency and independence from the mobile units to achieve their path-planning tasks. This paper proposes an improved grey wolf-optimized path-planning algorithm and a federated learning mechanism to improve the path-planning task efficiency and reduce resource consumption. A design solution for a network-switching and distributed communication facility is presented, then used as the basis for a digital twin-sensing network. Experimental results show that the hardware platform functioned in compliance with the actual task requirements, that the new heuristic path planning algorithm outperformed other algorithms in its class, and that the federated learning mechanism improved the parameter setting efficiency in the planning algorithm. The proposed model demonstrably improved the path-planning efficiency of mobile units at the Antarctic research stations. Moreover, a series of simulations and field experiments at Zhongshan Station confirmed that the proposed algorithm achieved good performance in global heuristic path planning, planning cost evaluation, and regional dynamic path planning.

Key words:

path planning, heuristic algorithms, Zhongshan Station, federated learning