|
作者简介: 丁国胜(1982—),男,硕士,研究方向为信息融合。 |
|
蔡民杰(1989—),男,硕士。 |
Copy editor: 张培培
收稿日期: 2021-11-18
要求修回日期: 2021-12-29
网络出版日期: 2022-04-28
版权
Multi-target Point-Track Association Method Based on Reinforcement Learning
Received date: 2021-11-18
Request revised date: 2021-12-29
Online published: 2022-04-28
Copyright
针对密集杂波环境下的多目标点迹-航迹关联问题,以强化学习(Reinforcement Learning, RL)方法为基础,提出了一种基于Q学习的多目标点迹-航迹关联方法。首先,根据整个过程中目标的运动状态,建立马尔可夫决策过程(Markov Decision Process, MDP)模型。其次,利用各状态间的相关程度构成策略函数,选择准确的动作,并设定相应的奖励函数。最后,考虑杂波密集时虚假量测难以分辨,结合目标先验信息,增加了Q表再学习环节,进一步优化关联精度。仿真结果表明,在非机动和强机动两种环境下,该方法都能准确地关联到目标的量测,具有较好的点迹-航迹关联性能。
关键词: 多目标点迹-航迹关联; 强化学习; MDP模型; 策略函数; Q表再学习
丁国胜 , 蔡民杰 . 基于强化学习的多目标点航关联方法[J]. 指挥控制与仿真, 2022 , 44(2) : 43 -48 . DOI: 10.3969/j.issn.1673-3819.2022.02.009
Aiminging at the problem of multi-target point-track association in dense clutter environment, based on the reinforcement learning(RL) method, a multi-target point-track association method based on Q-learning is proposed. First, according to the movement state of the target in the whole process, a Markov decision process(MDP) model is established. Secondly, the paper uses the degree of correlation between the states to form a strategy function, selects the correct action, and sets the corresponding reward function. Finally, considering that false measurements are difficult to distinguish when the clutter is dense, combined with the prior information of the target, the Q-meter re-learning link is added to further optimize the correlation accuracy. The simulation results show that in both non-maneuvering and strong maneuvering environments, the method in this paper can accurately correlate to the measurement of the target, and has a better point-track-track correlation performance.
| [1] |
黄小平, 王岩. 卡尔曼滤波原理及应用—MATLAB 仿真[M]. 北京: 电子工业出版社, 2015.
|
| [2] |
|
| [3] |
|
| [4] |
|
| [5] |
孙启臣, 郭伟震, 闫倩倩, 等. 一种基于灰关联分析的多目标跟踪算法[J]. 鲁东大学学报(自然科学版), 2017, 33(1): 20-25,97.
|
| [6] |
|
| [7] |
周志华. 机器学习[M]. 北京: 清华大学出版社, 2016.
|
| [8] |
|
| [9] |
|
| [10] |
|
| [11] |
|
| [12] |
|
| [13] |
|
| [14] |
董豪, 丁子涵. Deep Reinforcement Learaing[M]. 北京: 电子工业出版社, 2021.
|
/
| 〈 |
|
〉 |