-
摘要:
为准确感知机场场面运行环境，提出基于度量学习的交通态势弱监督评估方法。根据机场场面航空器的时空分布类型，从交通流量、起降队列、资源需求等视角构建交通态势指标体系；借鉴度量学习范式，利用预先定义的相似集和不相似集自动学习态势样本之间的距离度量；在此基础上，采用K均值算法实现弱监督条件下交通态势的等级划分。以上海浦东国际机场实际运行数据为例，分析并验证所提方法的有效性。实验结果表明：起始时刻离场瞬时流量、离场累计流量、离场跑道队列长度及进场累计流量的距离系数大于0.5，对场面态势影响较大；与基于欧式距离的K均值算法相比，度量学习将最优轮廓系数提升了33.3%，得到符合预期语义的聚类结果；此外，机场的平均滑行时间越长、跑道配置越复杂，其场面交通态势等级越高。
Abstract:In order to accurately perceive the operating environment of the airport surface, a weakly supervised evaluation method of traffic situation based on metric learning is proposed. Firstly, according to the space-time distribution types of aircraft on the airport surface, the traffic situation index system is constructed from the perspectives of traffic flow, take-off and landing queues, and resource demand; secondly, learning from the metric learning paradigm, the distance measure between situation samples is automatically learned by using pre-defined similar sets and dissimilar sets; on this basis, the K-means algorithm is used to achieve traffic situation classification under weak supervision. Taking the actual operation data of Shanghai Pudong International Airport as an example, the effectiveness of the proposed method is analyzed and verified. The experimental results show that:when the distance coefficients of the departure instantaneous flow at the beginning, the departure cumulative flow, the length of the departure runway queue, and the arrival cumulative flow are greater than 0.5, it has a greater impact on the surface situation; metric learning improves the optimal contour coefficient by 33.3% when compared to the K-means algorithm based on Euclidean distance, and obtains clustering results that meet the e-criterion; in addition, the longer the average taxi time of the airport and the more complex the runway configuration, the higher the level of the traffic situation on the surface.
-
Key words:
- air transportation /
- airport congestion /
- situation assessment /
- metric learning /
- clustering analysis
-
图 4 进离场航班不同跑道配置下平均滑行时间分布
Figure 4. Distribution of average taxi time under different runway configurations of departure and arrival flights
图 7 不同态势等级下平均滑行时间的概率密度分布
Figure 7. Probability density distribution of average taxi time under different situation levels
图 8 不同经验参数下最优轮廓系数
Figure 8. Optimal contour coefficient under different empirical parameters
表 1 交通态势指标体系
Table 1. Traffic situation index system
指标类型 指标名称 解释说明 离场类指标 起始时刻离场瞬时流量d1 参考时间片起始时刻处于离场过程的航空器架次 终止时刻离场瞬时流量d2 参考时间片终止时刻处于离场过程的航空器架次 离场累计流量d3 参考时间片内所有离场航班架次 离场跑道队列长度d4 参考时间片内从跑道起飞的航班架次 离场资源需求指数d5 参考时间片内从停机位推出的航班架次 进场类指标 起始时刻进场瞬时流量a1 参考时间片起始时刻处于进场过程的航空器架次 终止时刻进场瞬时流量a2 参考时间片终止时刻处于进场过程的航空器架次 进场累计流量a3 参考时间片内所有进场航班架次 进场跑道队列长度a4 参考时间片内在跑道降落的航班架次 进场资源需求指数a5 参考时间片内推入到停机位的航班架次 
下载: 导出CSV
表 2 指标的基本统计信息
Table 2. Basic statistics of indices
指标 统计量/架次 最小值 最大值 均值 中位数 d1 0 25 8.99 9 d2 0 23 8.92 9 d3 1 38 16.99 18 d4 0 17 8.07 9 d5 0 20 8.01 8 a1 0 18 7.53 8 a2 0 17 7.49 8 a3 0 31 15.67 16 a4 0 17 8.15 8 a5 0 18 8.18 8
下载: 导出CSV
表 3 度量学习算法学习到的距离矩阵的估计结果
Table 3. Estimation results of distance matrix learned by metric learning algorithm
指标 d1 d2 d3 d4 d5 a1 a2 a3 a4 a5 d1 0.50 −0.09 −0.18 −0.26 0.21 0.07 0.07 0.01 −0.07 −0.06 d2 −0.09 0.25 −0.22 0.20 −0.29 0.06 0.06 0.06 −0.06 −0.05 d3 −0.18 −0.22 0.77 −0.22 −0.21 0.02 0.02 0.04 −0.02 −0.02 d4 −0.26 0.20 −0.22 0.57 −0.08 −0.04 −0.04 −0.01 0.03 0.03 d5 0.21 −0.29 −0.21 −0.08 0.34 −0.05 −0.05 −0.02 0.05 0.04 a1 0.07 0.06 0.02 −0.04 −0.05 0.36 −0.16 −0.22 0.28 −0.23 a2 0.07 0.06 0.02 −0.04 −0.05 −0.16 0.32 −0.21 −0.21 0.29 a3 0.01 0.06 0.04 −0.01 −0.02 −0.22 −0.21 0.76 −0.21 −0.22 a4 −0.07 −0.06 −0.02 0.03 0.05 0.28 −0.21 −0.21 0.32 −0.17 a5 −0.06 −0.05 −0.02 0.03 0.04 −0.23 0.29 −0.22 −0.17 0.35
下载: 导出CSV
表 4 关键指标的聚类中心
Table 4. Cluster center of key indices
指标 聚类中心 态势等级1 态势等级2 态势等级3 d1 2.83 10.42 16.06 d3 3.37 10.26 16.48 d4 8.01 19.77 27.31 a3 10.65 15.49 20.14
下载: 导出CSV
表 5 不同态势等级下占比大于5%的跑道配置情况
Table 5. Runway configuration accounting for more than 5% under different situation levels
态势等级 跑道配置（进场|离场） 占比/% 1 34R|35R 37.5 35L|34L 26.2 17R|16R 10.7 16L|17L 8.0 34R,35L|35R 5.9 2 34R,35L|34L,35R 52.1 17R,16L|17L,16R 41.6 3 17R,16L|17L,16R 48.5 34R,35L|34L,35R 40.7
下载: 导出CSV
-
[1] SIMAIAKIS I, BALAKRISHNAN H. A queuing model of the airport departure process[J]. Transportation Science, 2016, 50(1): 94-109. doi: 10.1287/trsc.2015.0603 [2] MARIN A G. Airport management: Taxi planning[J]. Annals of Operations Research, 2006, 143(1): 191-202. doi: 10.1007/s10479-006-7381-2 [3] 郑丽君, 胡荣, 张军峰, 等. 高峰时段下离港航空器绿色滑行策略设计与评价[J]. 北京航空航天大学学报, 2019, 45(11): 2320-2326. doi: 10.13700/j.bh.1001-5965.2019.0124ZHENG L J, HU R, ZHANG J F, et al. Design and evaluation of green taxiing strategy for departure aircraft during peak hours[J]. Journal of Beijing University of Aeronautics and Astronautics, 2019, 45(11): 2320-2326(in Chinese). doi: 10.13700/j.bh.1001-5965.2019.0124 [4] RODRÍGUEZ-SANZ Á, COMENDADOR F G, VALDÉS R A, et al. Characterization and prediction of the airport operational saturation[J]. Journal of Air Transport Management, 2018, 69: 147-172. doi: 10.1016/j.jairtraman.2018.03.002 [5] SMELTINK J W, SOOMER M J, DE WAAL P R, et al. Optimisation of airport taxi planning: NLRTP-2003-475[R]. Amsterdam : National Aerospace Laboratory NLR, 2003. [6] 李善梅, 徐肖豪, 孟令航. 基于聚类神经网络的机场拥挤等级预测[J]. 计算机工程与应用, 2013, 49(17): 254-257. doi: 10.3778/j.issn.1002-8331.1205-0003LI S M, XU X H, MENG L H. Forecasting of airport congestion level based on cluster and neural network algorithms[J]. Computer Engineering and Applications, 2013, 49(17): 254-257(in Chinese). doi: 10.3778/j.issn.1002-8331.1205-0003 [7] 谢燕雯. 基于隐马尔可夫过程的机场运行态势预测方法研究[D]. 哈尔滨: 哈尔滨工业大学, 2018: 21-32.XIE Y W. Research on airport operation situation prediction method based on hidden Markov process[D]. Harbin: Harbin Institute of Technology, 2018: 21-32 (in Chinese). [8] BALAKRISHNA P, GANESAN R, SHERRY L. Airport taxi-out prediction using approximate dynamic programming: Intelligence-based paradigm[J]. Transportation Research Record, 2008, 2052(1): 54-61. doi: 10.3141/2052-07 [9] BALAKRISHNA P, GANESAN R, SHERRY L. Accuracy of reinforcement learning algorithms for predicting aircraft taxi-out times: A case-study of Tampa Bay departures[J]. Transportation Research Part C:Emerging Technologies, 2010, 18(6): 950-962. doi: 10.1016/j.trc.2010.03.003 [10] YIN J N, HU M H, MA Y Y, et al. Airport taxi situation awareness with a macroscopic distribution network analysis[J]. Networks and Spatial Economics, 2019, 19(3): 669-695. doi: 10.1007/s11067-018-9402-5 [11] CLEWLOW R, SIMAIAKIS I, BALAKRISHNAN H. Impact of arrivals on departure taxi operations at airports[C]//AIAA Guidance, Navigation, and Control Conference. Reston: AIAA, 2010: 7698. [12] XING E P, NG A Y, JORDAN M I, et al. Distance metric learning with application to clustering with side-information[C]//Proceedings of 15th International Conference on Neural Information Processing Systems. Cambridge: MIT Press, 2002, 521–528. [13] DU J H, HU M H, ZHANG W N. Decision support for aircraft taxi time based on deep metric learning[C]//2020 IEEE 23rd International Conference on Intelligent Transportation Systems. Piscataway: IEEE Press, 2020: 1-7. [14] LIU Y, SEELHORST M, POZDNUKHOV A, et al. Assessing terminal weather forecast similarity for strategic air traffic management [C]//6th International Conference on Research in Air Transportation. Istanbul: Istanbul Technical University, 2014: 26-30. [15] BOOTH D, FLYNN B, SULTANA J. Airport CDM implementation[EB/OL]. (2021-08-09)[2017-03-31]. https://www.eurocontrol.int/sites/default/files/publication/files/airport-cdm-manual-2017.PDF. [16] ROUSSEEUW P J. Silhouettes: A graphical aid to the interpretation and validation of cluster analysis[J]. Journal of Computational and Applied Mathematics, 1987, 20: 53-65. doi: 10.1016/0377-0427(87)90125-7 -
下载:
点击查看大图
计量
- 文章访问数: 247
- HTML全文浏览量: 70
- PDF下载量: 22
- 被引次数: 0
