基于REINFORCE算法的组网测向选站方法
A Station Selection Method for Networked Direction-Finding Based on the REINFORCE Algorithm
针对传统组网测向方法应对高速移动、短时出联目标信号时,存在时效性差、难以实时跟监的问题,提出一种基于REINFORCE算法的组网测向选站方法。采用马尔可夫决策过程(Markov Decision Process, MDP)框架,将站点组合过程分解为多步决策过程,通过逐次选取站点来构建组合,实现组合优化问题到序贯决策问题的转化;同时,以站点与目标的位置关系作为状态空间,用选站组合数量约束智能体的学习步数,对已选择的站点进行动作屏蔽,并利用几何精度因子(Geometrical Dilution of Precision, GDOP)作为奖惩函数评价选站方案的优劣。实验结果表明,该方法在时间复杂度上显著优于遍历方法与遗传算法,在定位精度上与遍历方法基本持平,在泛化能力上,能够学习到测向站空间构型与GDOP之间的映射关系,对于未训练的目标场景也可以完成组网方案决策。
To address the issue of the poor timeliness and difficulty of real-time tracking in traditional networked direction-finding methods when handling fast-moving targets with short signal burst durations, a station selection method based on the REINFORCE algorithm is proposed. A Markov decision process (MDP) framework is employed to decompose the station combination process into a multi-step sequential decision process. By selecting stations one by one, the combinatorial optimization problem is transformed into a sequential decision problem. The positional relationship between stations and the target is used as the state space, the number of stations in the combination constrains the agent’s learning steps, action masking is applied to the selected stations, and the geometrical dilution of precision(GDOP) serves as the reward function to evaluate the quality of station selection schemes. Experimental results demonstrate that the proposed method significantly outperforms both the exhaustive method and the genetic algorithm in terms of time complexity, while achieving comparable localization accuracy to the exhaustive method. Furthermore, it exhibits strong generalization ability by learning the mapping between station spatial configuration and GDOP, enabling effective network deployment decisions for unseen target scenarios.
| [1] |
刘宏建,王明孝,蔡中祥, |
| [2] |
张世航.基于TDOA的对地目标定位技术研究[D].成都:电子科技大学,2023:49. |
| [3] |
赵家凯.基于群智能算法的无源时差定位最优布站研究[D].西安:西安电子科技大学,2022:15. |
| [4] |
曹文焕,韩升明,陈超.多站测向交叉定位的定位精度研究[J].现代电子技术,2026,49(5):51-55. |
| [5] |
孟圣波,李鹏飞,黄建廷.对空观察哨双站测向交叉定位误差分析与验证[J].舰船电子工程,2023,43(7):36-39. |
| [6] |
窦雪倩,李华,许博.基于几何稀释度的多站外源时差定位最优布站方法[J].电子信息对抗技术,2018,33(5):37-40. |
| [7] |
曾凤,蔡方凯,龙宁.基于自适应DE算法的TDOA定位布站优化策略[J].电子信息对抗技术,2025,40(1):53-62. |
| [8] |
丁磊,任潞,侯轩, |
| [9] |
高铭壑.基于多目标融合的充电桩位置规划方法研究[D].北京:华北电力大学,2020:3. |
| [10] |
刘雅婷.基于多智能体强化学习的应急避难所动态选址算法研究[D].北京:北京交通大学,2023:7. |
| [11] |
|
| [12] |
张天予,滕鹏晓,吕君, |
| [13] |
肖秀丽,唱亮.测向交会定位的定位精度分析[J].中国无线电,2011(4):51-53;66. |
| [14] |
杨思明,单征,丁煜, |
| [15] |
祝亢,黄珍,王绪明.基于深度强化学习的智能船舶航迹跟踪控制[J].中国舰船研究,2021,16(1):105-113. |
| [16] |
万里鹏,兰旭光,张翰博, |
/
| 〈 |
|
〉 |