航空学报 > 2024, Vol. 45 Issue (4): 328735-328735   doi: 10.7527/S1000-6893.2023.28735

基于路径和算法的飞行器集群分布式相对定位

赖玮清, 万九卿()   

  1. 北京航空航天大学 自动化科学与电气工程学院,北京 100191
  • 收稿日期:2023-03-22 修回日期:2023-05-05 接受日期:2023-07-27 出版日期:2024-02-25 发布日期:2023-08-04
  • 通讯作者: 万九卿 E-mail:wanjiuqing@buaa.edu.cn
  • 基金资助:
    国家自然科学基金(61873015)

Distributed relative positioning of aircraft group based on path⁃sum algorithm

Weiqing LAI, Jiuqing WAN()   

  1. School of Automation Science and Electrical Engineering,Beihang University,Beijing 100191,China
  • Received:2023-03-22 Revised:2023-05-05 Accepted:2023-07-27 Online:2024-02-25 Published:2023-08-04
  • Contact: Jiuqing WAN E-mail:wanjiuqing@buaa.edu.cn
  • Supported by:
    National Natural Foundation of China(61873015)

摘要:

根据飞行器集群中各节点的惯性测量以及节点间数据链给出的相对距离测量,利用集中式卡尔曼滤波可以实现集群节点间的相对定位,包括相对位置和相对姿态的估计。集中式算法将集群内全部数据汇聚到中心节点集中处理,系统的可扩展性和可靠性受到制约。以信念传播为代表的分布式相对定位算法,通过集群各节点对局部信息的处理以及相邻结点间的信息交互,可实现节点间相对位置和相对姿态的全局估计。然而,在集群相对定位问题中,数据链拓扑包含大量环状结构,此时信念传播算法存在收敛性问题。本文提出一种集群相对定位分布式精确推理方法,可适用于任意集群拓扑结构。分别在高斯标准型和规范型描述下,设计集群各节点误差状态的分布式时间更新和量测更新算法,基于路径和原理给出标准型和规范型参数的分布式转换算法。在线性高斯模型假设下,本文方法等价于集中式卡尔曼滤波,可实现相对定位的最优估计。设计基于集群分解的分布式近似推理算法,进一步提升算法运行速度。在六自由度长航时仿真数据上的计算结果表明,基于路径和的分布式近似推理的相对定位精度与集中式卡尔曼滤波接近,明显优于现有的分布式高斯信念传播算法。

关键词: 分布式推理, 相对定位, 信念传播, 路径和, 导航系统

Abstract:

According to the inertial measurement of each node in the aircraft group and the relative distance measurement given by the data link between nodes, the relative positioning including the estimation of the relative position and relative attitude between node pairs can be realized by using the centralized Kalman filter. The centralized algorithm gathers all the data in the group into the central node for centralized processing, and the scalability and reliability of the system are restricted. The distributed relative positioning algorithm such as belief propagation can realize global estimation of the relative position and relative attitude through the processing of local information by each node of the cluster and the information interaction between adjacent nodes. However, in the problem of group relative positioning, the data link topology contains a large number of ring structures, and the belief propagation algorithm has a convergence problem in such case. This paper proposes a distributed exact inference method for relative positioning, which can be applied to any group topology. Under the Gaussian standard and canonical descriptions, the distributed time update and measurement update algorithms for the error states of each node are designed. The distributed algorithms for standard and normative parameters conversion are given based on paths-sum principles. Under the linear Gaussian model assumption, the method in this paper is equivalent to the centralized Kalman filter, which can achieve the optimal estimation of relative positioning. A distributed approximate inference algorithm is proposed based on group decomposition to improve the speed of the algorithm. Results on the six-degree-of-freedom long-duration simulation data show that the relative positioning accuracy of the distributed approximate inference based on the path-sum is close to that of the centralized Kalman filter, and significantly outperforms the existing distributed Gaussian belief propagation algorithm.

Key words: distributed inference, relative positioning, belief propagation, path-Sum, navigation system

中图分类号: