进港航班滑入时间预测

唐小卫; 丁叶; 张生润; 任思豫; 吴佳琦

doi:10.13700/j.bh.1001-5965.2022.0625

进港航班滑入时间预测

doi: 10.13700/j.bh.1001-5965.2022.0625

南京航空航天大学民航学院，南京 211106

基金项目: 国家自然科学基金(61603178,U2333204,U2233208)

详细信息

通讯作者:
E-mail：tangxiaowei@nuaa.edu.cn

中图分类号: V351
计量
- 文章访问数: 478
- HTML全文浏览量: 85
- PDF下载量: 20
- 被引次数: 0
出版历程
- 收稿日期: 2022-07-19
- 录用日期: 2022-08-19
- 网络出版日期: 2022-09-09
- 整期出版日期: 2024-07-18

Taxi-in time prediction of arrival flight

College of Civil Aviation，Nanjing University of Aeronautics and Astronautics，Nanjing 211106，China

Funds: National Natural Science Foundation of China (61603178,U2333204,U2233208)

More Information

Corresponding author: E-mail：tangxiaowei@nuaa.edu.cn

摘要

摘要:
准确预测进港航班滑入时间对合理调配航班保障资源和提高机场场面运行效率具有重要意义，可有效克服各大机场粗放式预测航班进港时刻的不足，为此提出一种基于机器学习模型的滑入时间预测方法。以首都机场为具体研究对象，分析进港航班滑入时间的影响因素并构建特征集；将线性回归、K-最近邻、支持向量机、决策树、随机森林和梯度提升回归树6种在滑出时间预测方面得到广泛应用的机器学习模型用于进港航班滑入时间预测。研究结果表明：在误差范围±3 min内6种机器学习模型的预测精度均超过90%，表明特征集的构建和模型的选择是有效的；综合预测性能与模型拟合评估结果，梯度提升回归树模型的预测效果最好；在梯度提升回归树模型上场面流量特征的贡献度最大，新引入的跨区特征对预测模型的贡献度超过了大部分传统特征。
- 航空运输 /
- 机场场面运行 /
- 滑行时间预测 /
- 机器学习 /
- 梯度提升回归树
Abstract:
Accurate prediction of flight taxi-in time has a significant meaning in allocating aircraft support resources reasonably and improving airport surface movement efficiency. Therefore, a method of taxi-in time prediction based on machine learning model is proposed. It can effectively overcome the deficiency of extensive aircraft arrival time prediction in major airports currently. Using Beijing Capital International Airport as the research object, we firstly analyzed the factors that influence the taxi-in time and created the feature set. Next, we applied various techniques that are commonly used to predict taxi-out times, such as linear regression, K-nearest neighbor, support vector regression, decision tree, random forest, and gradient boosting regression tree, to predict the taxi-in time. The results show that the prediction accuracy of the six machine learning models is over 90% within ±3 min, which means that the construction of the feature set and the selection of models are effective. The gradient boosting regression tree model has the best performance based on the prediction results and model fitting evaluation results. The prediction results of gradient boosting regression tree show that the surface traffic flow features contribute most to the prediction model, and the newly proposed cross-regional feature contributes more than most traditional features.
- air transportation /
- airport surface movement /
- taxi time prediction /
- machine learning /
- gradient boosting regression tree

HTML全文

图 1 首都机场场面滑行流向（向北运行）

Figure 1. Taxiing direction on the surface of Capital Airport field (running north)

下载: 全尺寸图片幻灯片

图 2 场面运行航空器类别

Figure 2. Category of aircraft operation aircrafts on the surface

下载: 全尺寸图片幻灯片

图 3 6种模型的学习曲线

Figure 3. Learning curves for six models

下载: 全尺寸图片幻灯片

表 1 进港滑入时间统计数据

Table 1. Statistical data of taxi-in time

统计指标	进港滑入时间/min	统计指标	进港滑入时间/min
最小值	3	最大值	59
25%位数	7	平均值	12.35
中位数	11	标准差	6.16
75%位数	17

下载: 导出CSV

表 2 场面流量特征定义

Table 2. Definition of surface traffic flow features

特征变量	定义
$ {A_1} $	$ \begin{gathered} {A_1}\left( i \right) = {\mathrm{count}}\left( j \right),{t_{{\text{ALDT}}}}\left( j \right) \lt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{ALDT}}}}\left( i \right) \lt {t_{{\text{AIBT}}}}\left( j \right) \lt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {A_2} $	$ \begin{gathered} {A_2}\left( i \right) = {\mathrm{count}}\left( j \right),{t_{{\text{ALDT}}}}\left( j \right) \lt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{AIBT}}}}\left( j \right) \gt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {A_3} $	$ \begin{gathered} {A_3}\left( i \right) = {\mathrm{count}}\left( j \right),{t_{{\text{ALDT}}}}\left( j \right) \gt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{AIBT}}}}\left( j \right) \lt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {A_4} $	$ \begin{gathered} {A_4}\left( i \right) = {\mathrm{count}}\left( j \right),{t_{{\text{ALDT}}}}\left( i \right) \lt {t_{{\text{ALDT}}}}\left( j \right) \lt {t_{{\text{AIBT}}}}\left( i \right)\& \\ {t_{{\text{AIBT}}}}\left( j \right) \gt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {D_1} $	$ \begin{gathered} {D_1}\left( i \right) = {\mathrm{count}}\left( k \right),{t_{{\text{AOBT}}}}\left( k \right) \lt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{ALDT}}}}\left( i \right) \lt {t_{{\text{ATOT}}}}\left( k \right) \lt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {D_2} $	$ \begin{gathered} {D_2}\left( i \right) = {\mathrm{count}}\left( k \right),{t_{{\text{AOBT}}}}\left( k \right) \lt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{ATOT}}}}\left( k \right) \gt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {D_3} $	$ \begin{gathered} {D_3}\left( i \right) = {\mathrm{count}}\left( k \right),{t_{{\text{AOBT}}}}\left( k \right) \gt {t_{{\text{ALDT}}}}\left( i \right)\& \\ {t_{{\text{ATOT}}}}\left( k \right) \lt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
$ {D_4} $	$ \begin{gathered} {D_4}\left( i \right) = {\mathrm{count}}\left( k \right),{t_{{\text{ALDT}}}}\left( i \right) \lt {t_{{\text{AOBT}}}}\left( k \right) \lt {t_{{\text{AIBT}}}}\left( i \right)\& \\ {t_{{\text{ATOT}}}}\left( k \right) \gt {t_{{\text{AIBT}}}}\left( i \right) \\ \end{gathered} $
注：$ {t_{{\text{AOBT}}}} $为航班实际撤轮挡时刻，$ {t_{{\text{ATOT}}}} $为航班实际起飞离地时刻。

下载: 导出CSV

表 3 场面流量与滑入时间的相关性

Table 3. The correlation coefficient of surface traffic flow and taxi-in time

特征变量	相关系数	特征变量	相关系数
$ {A_1} $	0.573^***	$ {D_1} $	0.701^***
$ {A_2} $	−0.564^***	$ {D_2} $	−0.451^***
$ {A_3} $	0.878^***	$ {D_3} $	0.662^***
$ {A_4} $	0.662^***	$ {D_4} $	0.660^***
注：^***表示1%的显著性水平。

下载: 导出CSV

表 4 跨区与不跨区航班对应的平均滑入时间

Table 4. The average taxi-in time of cross-regional flights and non-cross-regional flights

是否跨区	样本量/%	平均滑行时长/min
不跨区	71	9.3
跨区	29	19.8

下载: 导出CSV

表 5 用于滑入时间预测的特征集构建

Table 5. Construction of feature set for taxi-in time prediction

特征名称	特征代号	特征描述
场面流量	$ \begin{array}{l}{A}_{1},{A}_{2},{A}_{3},{A}_{4}、\\ {D}_{1},{D}_{2},{D}_{3},{D}_{4}\end{array} $	见表2
跨区运行	$ K $	跨区为1，不跨区为0
滑行距离	$ d $	跑道口至停机位的滑行距离
进港跑道	$ W $	3条跑道共6端，跑道号为19、01、18L、36R、18R、36L的跑道端分别对应1～6编号
运行时段	$ T $	全天聚类为[0:00—0:59,10:00—23:59]、[1:00—1:59,9:00—9:59]、[2:00—8:59]共3个时段，分别对应1～3编号
机型	$ M $	机型共分为C、D、E、F四类，分别对应1～4编号
航司属性	$ I $	国内航司为1，国外航司为0

下载: 导出CSV

表 6 滑出时间预测研究中6种模型的预测精度

Table 6. Prediction accuracy of six models in the research of taxi-out time prediction

模型	R²	预测精度/%
模型	R²	±3 min	±5 min
LR^[11]	0.81	73.74	88.40
SVR^[13]	0.80	75.40	87.40
KNN^[14]	0.79	59.69	81.46
DT^[14]	0.81	63.07	85.10
RF^[11]	0.84	77.79	90.53
GBRT^[11]	0.83	77.36	90.19

下载: 导出CSV

表 7 6种模型的滑入时间预测性能评估

Table 7. Evaluation of taxi-in time prediction performance by six models

模型	S_RMSE/min	S_MAE/min	R²	预测精度/%
模型	S_RMSE/min	S_MAE/min	R²	±1 min	±2 min	±3 min
LR^[11]	2.01	1.39	0.89	49.6	78.3	90.3
SVR^[13]	1.63	1.12	0.93	58.4	85.7	94.3
KNN^[14]	1.85	1.26	0.91	56.4	82.3	92.3
DT^[14]	1.73	1.22	0.92	54.5	82.5	93.0
RF^[11]	1.50	1.07	0.94	59.6	86.5	95.0
GBRT^[11]	1.40	1.01	0.95	61.4	88.4	96.0

下载: 导出CSV

表 8 GBRT模型的特征重要度排序

Table 8. Feature importance order of GBRT model

排序	特征	特征重要度	排序	特征	特征重要度
1	$ {A_3} $	0.636 5	8	$ {D_4} $	0.013 6
2	$ d $	0.121 0	9	$ {D_2} $	0.008 1
3	$ K $	0.055 0	10	$ T $	0.003 6
4	$ {D_3} $	0.050 9	11	$ W $	0.003 3
5	$ {A_4} $	0.048 6	12	$ {A_1} $	0.001 4
6	$ {D_1} $	0.040 3	13	$ M $	0.000 6
7	$ {A_2} $	0.017 0	14	$ I $	0.000 1

下载: 导出CSV

参考文献(17)

[1]	HEBERT J E, DIETZ D C. Modeling and analysis of an airport departure process[J]. Journal of Aircraft, 1997, 34(1): 43-47. doi: 10.2514/2.2133
[2]	IDRIS H, CLARKE J P, BHUVA R, et al. Queuing model for taxi-out time estimation[J]. Air Traffic Control Quarterly, 2002, 10(1): 1-22. doi: 10.2514/atcq.10.1.1
[3]	RAVIZZA S, ATKIN J A D, MAATHUIS M H, et al. A combined statistical approach and ground movement model for improving taxi time estimations at airports[J]. Journal of the Operational Research Society, 2013, 64(9): 1347-1360. doi: 10.1057/jors.2012.123
[4]	JORDAN R, ISHUTKINA M A, REYNOLDS T G. A statistical learning approach to the modeling of aircraft taxi time[C]// 29th Digital Avionics Systems Conference. Piscataway: IEEE Press, 2010.
[5]	LEE H, MALIK W, JUNG Y C. Taxi-out time prediction for departures at charlotte airport using machine learning techniques[C]// Proceedings of the 16th AIAA Aviation Technology, Integration, and Operations Conference. Reston, Virginia: AIAA, 2016.
[6]	HERREMA F, CURRAN R, VISSER H, et al. Taxi-out time prediction model at Charles de Gaulle Airport[J]. Journal of Aerospace Information Systems, 2018, 15(3): 120-130. doi: 10.2514/1.I010502
[7]	冯霞, 孟金双. 基于KNN和SVR的航班滑出时间预测[J]. 西南交通大学学报, 2017, 52(5): 1008-1014. doi: 10.3969/j.issn.0258-2724.2017.05.023 FENG X, MENG J S. Flight taxi-out time prediction based on KNN and SVR[J]. Journal of Southwest Jiaotong University, 2017, 52(5): 1008-1014(in Chinese). doi: 10.3969/j.issn.0258-2724.2017.05.023
[8]	DIANA T. Can machines learn how to forecast taxi-out time? A comparison of predictive models applied to the case of Seattle/Tacoma International Airport[J]. Transportation Research Part E: Logistics and Transportation Review, 2018, 119: 149-164. doi: 10.1016/j.tre.2018.10.003
[9]	LIAN G, ZHANG Y P, XING Z W, et al. RETRACTED: a new dynamic pushback control method for reducing fuel-burn costs: using predicted taxi-out time[J]. Chinese Journal of Aeronautics, 2019, 32(3): 660-673. doi: 10.1016/j.cja.2018.12.013
[10]	李楠, 焦庆宇, 张连东, 等. 离场航空器滑行时间预测研究[J]. 重庆交通大学学报(自然科学版), 2021, 40(3): 1-6. LI N, JIAO Q Y, ZHANG L D, et al. Taxi time prediction of departure aircraft[J]. Journal of Chongqing Jiaotong University (Natural Science), 2021, 40(3): 1-6(in Chinese).
[11]	WANG X W, BROWNLEE A E I, WOODWARD J R, et al. Aircraft taxi time prediction: feature importance and their implications[J]. Transportation Research Part C: Emerging Technologies, 2021, 124: 102892. doi: 10.1016/j.trc.2020.102892
[12]	RAVIZZA S, CHEN J, ATKIN J A D, et al. Aircraft taxi time prediction: comparisons and insights[J]. Applied Soft Computing, 2014, 14: 397-406. doi: 10.1016/j.asoc.2013.10.004
[13]	LI N, JIAO Q Y, ZHU X H, et al. Prediction of departure aircraft taxi time based on deep learning[J]. Transactions of Nanjing University of Aeronautics and Astronautics, 2020, 37(2): 232-241.
[14]	陈祯, 唐小卫, 林有超, 等. 基于决策树的航空器离港滑出时间预测方法和模型[J]. 武汉理工大学学报(交通科学与工程版), 2021, 45(3): 448-453. CHEN Z, TANG X W, LIN Y C, et al. Prediction method and model of aircraft taxi-out time based on decision tree[J]. Journal of Wuhan University of Technology (Transportation Science & Engineering), 2021, 45(3): 448-453(in Chinese).
[15]	赵征, 冯事成, 宋梅雯, 等. 基于XGBoost的航空器动态滑行时间预测方法研究[J]. 航空工程进展, 2022, 13(1): 76-85. ZHAO Z, FENG S C, SONG M W, et al. Prediction method of aircraft dynamic taxi time based on XGBoost[J]. Advances in Aeronautical Science and Engineering, 2022, 13(1): 76-85(in Chinese).
[16]	BROWNLEE A E I, WEISZER M, CHEN J, et al. A fuzzy approach to addressing uncertainty in Airport Ground Movement optimisation[J]. Transportation Research Part C: Emerging Technologies, 2018, 92: 150-175. doi: 10.1016/j.trc.2018.04.020
[17]	沈志远, 胡莹莹. 考虑尾流影响的侧向双跑道机场的跑道容量研究[J]. 南京航空航天大学学报, 2020, 52(1): 161-170. SHEN Z Y, HU Y Y. Runway capacity of lateral double-runway airport considering wake effect[J]. Journal of Nanjing University of Aeronautics & Astronautics, 2020, 52(1): 161-170(in Chinese).