基于Trans-Attention的飞行区航空器监视数据融合方法

王兴隆; 尹昊; 丁俊峰

doi:10.13700/j.bh.1001-5965.2023.0234

基于Trans-Attention的飞行区航空器监视数据融合方法

doi: 10.13700/j.bh.1001-5965.2023.0234

1.
中国民航大学飞联网重点实验室，天津 300300
2.
南京莱斯信息技术股份有限公司，南京 210001

基金项目: 国家自然科学基金面上项目(62173332)；国家自然科学基金重点项目(U2133207)；天津多元基金项目(21JCYBJCO0700)；中国民航大学校级研究生科研创新项目(2022YJS086)

详细信息

通讯作者:
E-mail：xinglong1979@163.com

中图分类号: V351
计量
- 文章访问数: 30
- HTML全文浏览量: 9
- PDF下载量: 1
- 被引次数: 0
出版历程
- 收稿日期: 2023-05-09
- 录用日期: 2023-07-14
- 网络出版日期: 2023-08-30
- 整期出版日期: 2025-04-30

Aircraft surveillance data fusion method in flight area based on Trans-Attention

1.
Key Laboratory of Internet of Aircrafts，Civil Aviation University of China，Tianjin 300300，China
2.
Nanjing Les Information Technology Co.，Ltd.，Nanjing 210001，China

Funds: General Program of the National Natural Science Foundation of China (62173332); Key Program of the National Natural Science Foundation of China (U2133207); Diversified Fund Project of Tianjin, China (21JCYBJCO0700); School Level Graduate Research Innovation Project of Civil Aviation University of China (2022YJS086)

More Information

Corresponding author: E-mail：xinglong1979@163.com

摘要

摘要:
针对飞行区航空器单一监视源存在监视精度低、位置跳变的问题，提出一种基于Transformer和注意力机制的航空器监视数据融合方法。利用Transformer的编码器结构分别对各监视源数据进行特征提取，通过注意力机制对不同监视源赋予权重值，经过全连接网络进行回归计算，获得最终的融合结果。选取场面监视雷达(SMR)和广播式自动相关监视(ADS-B)系统的监视数据作为融合源，多点定位(MLAT)数据作为真实标签，实验结果表明：所提方法有效降低了单一监视源的监视误差，且融合效果优于基于注意力机制的长短期记忆网络、循环神经网络和扩展卡尔曼滤波融合方法，平均绝对误差分别提升了2.81%、16.73%和35.80%。
- 数据融合 /
- Transformer /
- 注意力机制 /
- 场面监视雷达 /
- 广播式自动相关监视
Abstract:
An aircraft surveillance data fusion method based on a Transformer and attention mechanism is proposed to address the issues of low monitoring accuracy and position jump in a single surveillance source for aircraft in the flight area. Prior to assigning weight values to various surveillance sources via the attention mechanism, features are first extracted from each surveillance source data using the Transformer’s encoder structure. Finally, regression calculations are performed through a fully connected network to obtain the final fusion result. The multilateration (MLAT) data are employed as actual tags, while the surveillance data from the autonomous dependent surveillance-broadcast (ADS-B) system and the surface movement radar (SMR) are chosen as fusion sources. The experimental results show that the proposed method effectively reduces the surveillance error of a single surveillance source, and the fusion effect is better than that of the long short-term memory network based on the attention mechanism, recurrent neural network, and extended Kalman filter fusion methods. The mean absolute error is increased by 2.81%, 16.73% and 35.80% respectively.
- data fusion /
- Transformer /
- attention mechanism /
- surface movement radar /
- automatic dependent surveillance-broadcast

HTML全文

图 1 时间对齐示意图

Figure 1. Time alignment diagram

下载: 全尺寸图片幻灯片

图 2 Trans-Attention融合模型结构

Figure 2. Trans-Attention fusion model structure

下载: 全尺寸图片幻灯片

图 3 多头注意力机制结构

Figure 3. Multi-head attention mechanism structure

下载: 全尺寸图片幻灯片

图 4 航空器直线滑行时模型融合结果

Figure 4. Model fusion results during aircraft straight-line taxiing

下载: 全尺寸图片幻灯片

图 5 航空器转弯滑行时模型融合结果

Figure 5. Model fusion results during aircraft turning and taxiing

下载: 全尺寸图片幻灯片

图 6 不同融合方法结果对比

Figure 6. Comparison of results of different fusion methods

下载: 全尺寸图片幻灯片

表 1 Trans-Attention融合模型参数设置

Table 1. Trans-Attention fusion model parameter settings

学习率	隐藏层丢失率	训练轮次	多头注意力个数	Encode层深度	批量大小
0.001	0.1	600	6	2 048	8

下载: 导出CSV

表 2 基于Trans-Attention融合方法结果数据评估

Table 2. Data evaluation based on Trans-Attention fusion method results

数据来源	MSE/m	RMSE/m	MAE/m
Trans-Attention	4.2811	2.0691	2.0595
ADS-B	14.4971	3.8075	4.3479
SMR1	36.3145	6.0262	8.2769
SMR2	47.4476	6.8882	10.6579

下载: 导出CSV

表 3 EKF融合方法参数设置

Table 3. EKF fusion method parameter settings

参数类型	参数值
初始化状态协方差矩阵 ${{\boldsymbol{P}}_0}$	$\left[ {\begin{array}{*{20}{c}} {{{1.0}}}&{{{0}}}&{{{0}}}&{{{0}}} \\ {{{0}}}&{{{1.0}}}&{{{0}}}&{{{0}}} \\ {{{0}}}&{{{0}}}&{{{1.0}}}&{{{0}}} \\ {{{0}}}&{{{0}}}&{{{0}}}&{{{1.0}}} \end{array}} \right]$
过程噪声协方差 ${\boldsymbol{Q}}_0$	$\left[ {\begin{array}{*{20}{c}} {0.000\;1}&{0}&{0}&{0}&{0} \\ {0}&{0.000\;1}&{0}&{0}&{0} \\ {0}&{0}&{0.000\;09}&{0}&{0} \\ {0}&{0}&{0}&{0.001}&{0} \\ {0}&{0}&{0}&{0}&{0.001} \end{array}} \right]$
SMR1的测量协方差矩阵 ${{\boldsymbol{R}}_1}$	$\left[ {\begin{array}{*{20}{c}} {6.25}&{0}&{0} \\ {0}&{0.001\;6}&{0} \\ {0}&{0}&{6.25} \end{array}} \right]$
SMR2的测量协方差矩阵 ${{\boldsymbol{R}}_2}$	$\left[ {\begin{array}{*{20}{c}} {4.622\;5}&{0}&{0} \\ {0}&{0.000\;9}&{0} \\ {0}&{0}&{4.625\;5} \end{array}} \right]$
ADS-B的测量协方差矩阵 ${{\boldsymbol{R}}_3}$	$\left[ {\begin{array}{*{20}{c}} {2.25}&{0} \\ {0}&{2.25} \end{array}} \right]$

下载: 导出CSV

表 4 RNN融合方法参数设置

Table 4. RNN fusion method parameter settings

学习率	隐藏层丢失率	训练轮次	批量大小	RNN层数
0.001	0.1	300	8	2

下载: 导出CSV

表 5 LSTM-Attention融合方法参数设置

Table 5. LSTM-Attention fusion method parameter settings

学习率	隐藏层丢失率	训练轮次	批量大小	LSTM层数
0.001	0.1	300	8	2

下载: 导出CSV

表 6 不同融合方法结果数据评估

Table 6. Data evaluation form for results of different fusion methods

方法	MSE/m	RMSE/m	MAE/m
本文方法	4.2811	2.0691	2.0595
EKF	9.9745	3.1582	3.2078
LSTM-Attention	4.3775	2.0922	2.1190
RNN	4.9697	2.2293	2.4732

下载: 导出CSV

参考文献(21)

[1]	LANZKRON P, BROOKNER E. Solid state X-band airport surface surveillance radar[C]//Proceedings of the IEEE Radar Conference. Piscataway: IEEE Press, 2007: 670-676.
[2]	卿烈华. 场面监视雷达系统在民用机场的应用[J]. 数字技术与应用, 2018, 36(5): 100-102. QING L H. Application of scene surveillance radar system in civil airports[J]. Digital Technology and Application, 2018, 36(5): 100-102(in Chinese).
[3]	王思达, 段昌淼. 基于ADS-B的场面监视技术探讨[J]. 空运商务, 2018(6): 59-61. doi: 10.3969/j.issn.1671-3095.2018.06.030 WANG S D, DUAN C M. Discussion on scene monitoring technology based on ADS-B[J]. Air Transport & Business, 2018(6): 59-61(in Chinese). doi: 10.3969/j.issn.1671-3095.2018.06.030
[4]	VASYLIEV V M, VASYLIEV D V, NAUMENKO K V. Using data of multilateration surveillance system for aircraft tracking[C]// Proceedings of the 4th International Conference on Methods and Systems of Navigation and Motion Control. Piscataway: IEEE Press, 2016: 279-283.
[5]	戴敏. 结合正则化求解的机场场面多点定位方法[J]. 计算机工程与设计, 2021, 42(10): 2995-3001. DAI M. Multilateration method for airport surface surveillance based on regularization[J]. Computer Engineering and Design, 2021, 42(10): 2995-3001(in Chinese).
[6]	AKRAM M A, LIU P L, TAHIR M O, et al. A state optimization model based on Kalman filtering and robust estimation theory for fusion of multi-source information in highly non-linear systems[J]. Sensors, 2019, 19(7): 1687. doi: 10.3390/s19071687
[7]	张学军, 张其善. ATM中的ADS-SSR数据融合研究[J]. 北京航空航天大学学报, 2001, 27(1): 24-27. doi: 10.3969/j.issn.1001-5965.2001.01.007 ZHANG X J, ZHANG Q S. Data fusion of ADS-SSR in air traffic management[J]. Journal of Beijing University of Aeronautics and Astronautics, 2001, 27(1): 24-27(in Chinese). doi: 10.3969/j.issn.1001-5965.2001.01.007
[8]	XIA X, HASHEMI E, XIONG L, et al. Autonomous vehicle kinematics and dynamics synthesis for sideslip angle estimation based on consensus Kalman filter[J]. IEEE Transactions on Control Systems Technology, 2022, 31(1): 179-192.
[9]	刘康, 何明浩, 韩俊, 等. 基于多传感器的雷达对抗侦察数据融合算法[J]. 系统工程与电子技术, 2023, 45(1): 101-107. LIU K, HE M H, HAN J, et al. Data fusion algorithm for radar countermeasures and reconnaissance based on multi-sensor[J]. Systems Engineering and Electronics, 2023, 45(1): 101-107(in Chinese).
[10]	李素. 基于航迹质量分析的加权平均融合算法[J]. 现代计算机(专业版), 2018, 24(5): 12-15. LI S. A weighted average fusion algorithm based on track quality analysis[J]. Modern Computer, 2018, 24(5): 12-15(in Chinese).
[11]	RAJASEKARAN R K, AHMED N, FREW E. Bayesian fusion of unlabeled vision and RF data for aerial tracking of ground targets[C]//Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems. Piscataway: IEEE Press, 2020: 1629-1636.
[12]	郭建刚, 陈鹏, 郑伟. 多表冗余惯导数据融合算法及在自对准中的应用[J]. 北京航空航天大学学报, 2020, 46(12): 2211-2216. GUO J G, CHEN P, ZHENG W. Data fusion algorithm of multi-sensor redundant inertial navigation and its application in self-alignment[J]. Journal of Beijing University of Aeronautics and Astronautics, 2020, 46(12): 2211-2216(in Chinese).
[13]	周恩帆, 马俊, 周永杰, 等. 一种D-S证据理论的多传感器数据融合算法[J]. 小型微型计算机系统, 2022, 43(4): 795-800. ZHOU E F, MA J, ZHOU Y J, et al. Multi-sensor data fusion algorithm based on D-S evidence theory[J]. Journal of Chinese Computer Systems, 2022, 43(4): 795-800(in Chinese).
[14]	KACZMAREK A, ROHM W, KLINGBEIL L, et al. Experimental 2D extended Kalman filter sensor fusion for low-cost GNSS/IMU/Odometers precise positioning system[J]. Measurement, 2022, 193: 110963. doi: 10.1016/j.measurement.2022.110963
[15]	宋安宇. 基于RNN的航空监视信息融合技术的研究与实现[D]. 北京: 北京邮电大学, 2019. SONG A Y. Research and implementation of aviation surveillance information fusion technology based on RNN[D]. Beijing: Beijing University of Posts and Telecommunications, 2019(in Chinese).
[16]	孟致远. 基于多神经网络的航空监视信息融合系统的研究与实现[D]. 北京: 北京邮电大学, 2021. MENG Z Y. The research and implementation of aviation surveillance information fusion system based on multi neural network[D]. Beijing: Beijing University of Posts and Telecommunications, 2021 (in Chinese).
[17]	陈建军, 孙俊, 杨予昊, 等. 机场场面监视雷达系统设计[J]. 现代雷达, 2018, 40(12): 11-14. CHEN J J, SUN J, YANG Y H, et al. Design of airport surface movement radar system[J]. Modern Radar, 2018, 40(12): 11-14(in Chinese).
[18]	LU Y, HUANG R S, XU Z L. Multi-sensor data fusion based on ADS-B and MLAT in approach[J]. Applied Mechanics and Materials, 2014, 602-605: 2491-2494. doi: 10.4028/www.scientific.net/AMM.602-605.2491
[19]	闫峰. 浅析北京大兴国际机场多点监视系统[J]. 电脑知识与技术, 2021, 17(21): 136-138. YAN F. Analysis of Beijing daxing international airport multipoint surveillance system[J]. Computer Knowledge and Technology, 2021, 17(21): 136-138(in Chinese).
[20]	牛中伟, 陈健明. 空管场监系统雷达数据格式分析[J]. 信息通信, 2019, 32(1): 156-158. NIU Z W, CHEN J M. Analysis of radar data format for air traffic control field supervision system[J]. Information & Communications, 2019, 32(1): 156-158(in Chinese).
[21]	VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30: 5998-6008.