航空学报 > 2016, Vol. 37 Issue (5): 1634-1643   doi: 10.7527/S1000-6893.2015.0321

面向再入目标跟踪的估计与辨识联合优化算法

张金凤, 何重阳, 梁彦   

  1. 西北工业大学 自动化学院, 西安 710072
  • 收稿日期:2015-07-08 修回日期:2015-11-26 出版日期:2016-05-15 发布日期:2015-12-24
  • 通讯作者: 梁彦,Tel.:029-88431308 E-mail:liangyan@nwpu.edu.cn E-mail:liangyan@nwpu.edu.cn
  • 作者简介:张金凤,女,硕士研究生。主要研究方向:目标跟踪,信息融合。Tel:029-88431306 E-mail:zhangjinfengzdh@mail.nwpu.edu.cn;何重阳,男,硕士研究生。主要研究方向:目标跟踪,信息融合。Tel:029-88431306 E-mail:hechongyang@mail.nwpu.edu.cn;梁彦,男,博士,教授,博士生导师。主要研究方向:估计理论,目标跟踪与识别。Tel:029-88431308 E-mail:liangyan@nwpu.edu.cn
  • 基金资助:

    国家自然科学基金(61135001,61374023,61374159);航空科学基金(20125153)

Joint optimization algorithm of estimation and identification for reentry target tracking

ZHANG Jinfeng, HE Chongyang, LIANG Yan   

  1. School of Automation, Northwestern Polytechnical University, Xi'an 710072, China
  • Received:2015-07-08 Revised:2015-11-26 Online:2016-05-15 Published:2015-12-24
  • Supported by:

    National Natural Science Foundation of China (61135001,61374023,61374159);Aeronautical Science Foundation of China (20125153)

摘要:

准确的弹道系数辨识和精确的目标状态估计是再入目标高精度跟踪与高可靠识别的关键。一方面,状态估计的误差会造成模型参数(弹道系数)的辨识风险;另一方面,模型参数的辨识偏差又会导致模型失配从而降低目标状态的估计精度。因此,需要实现再入目标的状态估计和参数辨识的联合优化。针对再入目标弹道系数未知情形,提出了一种基于期望最大化(EM)框架并采用粒子滤波(PF)平滑器实现的PF-EM联合优化算法。在E步基于粒子平滑器得到目标状态的后验平滑估计,M步采用数值优化算法更新上一次迭代的弹道系数,通过E步和M步的不断迭代,以保证状态估计和弹道系数辨识的一致性。算法仿真对比表明:所提算法的状态估计和参数辨识精度均优于传统的状态增广算法。

关键词: 目标跟踪, 再入目标, 弹道系数, 期望最大化(EM), 联合优化

Abstract:

Reliable identification of ballistic coefficient and accurate estimation of target state are important issues and coupled:the state estimation error may trigger identification risk while identification risk causes state estimation error due to modeling mismatch. Therefore, it is essential to estimate the target state and identify unknown model parameters jointly. In this paper, the joint optimization algorithm PF-EM is proposed for tracking a reentry target with unknown ballistic coefficient, which is realized by using particle filter (PF) smoother under the expectation-maximization (EM) iterative framework. In the E-step, the random particle sampling strategy is utilized to approximate the likelihood function to deal with the inherited nonlinearity. In the M-step, the numerical optimization algorithm is applied to update mass-to-drag ratio. In the simulation compared with the traditional algorithm which augments the state vector with the unknown parameter, the proposed algorithm shows the improvement in both state estimate and parameter identification.

Key words: target tracking, reentry target, ballistic coefficient, expectation-maximization (EM), joint optimization

中图分类号: