基于时空图卷积神经网络的离港航班延误预测

姜雨; 陈名扬; 袁琪; 戴垚宇

doi:10.13700/j.bh.1001-5965.2021.0415

基于时空图卷积神经网络的离港航班延误预测

doi: 10.13700/j.bh.1001-5965.2021.0415

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(U1933118,U2033205,71971114)

详细信息

通讯作者:
E-mail：jiangyu07@nuaa.edu.cn

中图分类号: V351
计量
- 文章访问数: 351
- HTML全文浏览量: 65
- PDF下载量: 72
- 被引次数: 0
出版历程
- 收稿日期: 2021-07-22
- 录用日期: 2021-10-15
- 网络出版日期: 2021-10-28
- 整期出版日期: 2023-05-31

Departure flight delay prediction based on spatio-temporal graph convolutional networks

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

Funds: National Natural Science Foundation of China (U1933118,U2033205,71971114)

More Information

Corresponding author: E-mail：jiangyu07@nuaa.edu.cn

摘要

摘要:
对于日益频发的机场航班延误，精准的航班延误预测是最重要的防范措施之一。通过谱图卷积将机场网络从不规则的图结构转换为规则的网络结构，利用图卷积神经网络（GCN）和门控线性单元（GLU）捕获网络中的时空相关性并组成时空卷积块，提出可以预测离港航班延误状况的时空图卷积神经网络（STGCN）。遴选美国51座枢纽机场构建机场网络，并预测未来一段时间内的机场离港准点率以检验STGCN用于预测航班延误的可行性。结果表明：当预测窗口为1天时，STGCN预测结果的平均绝对误差（MAE）相对于历史平均（HA）法、长短期记忆循环神经网络（LSTM）、堆栈自编码器（SAEs）分别下降了18.19%、10.45%、6.24%；当预测窗口为2天时，MAE分别下降了9.93%、3.96%、4.37%；当预测窗口为3天时，MAE分别下降了7.02%、2.47%、9.20%。实例证明STGCN相比传统模型能够显著提升航班延误预测的精度，并为机场制定延误决策提供参考指导。
- 航班延误预测 /
- 深度学习 /
- 机场网络 /
- 图卷积神经网络 /
- 门控线性单元
Abstract:
Accurate flight delay prediction is one of the most important preventive measures for the increasingly frequent airport flight delays. The airport network is transformed from irregular topological structure to regular network structure by spectral convolution. The graph convolutional network (GCN) and gated linear unit (GLU) are used to capture spatio-temporal correlation in the network and form spatio-temporal convolutional blocks. A spatio-temporal graph convolutional neural network (STGCN) is proposed to predict the departure flight delay. 51 major airports in the United States are selected to construct the airport network, and the on-time departure rate is predicted to carry out the example verification. The results show that when the prediction window is 1 day, the mean absolute error (MAE) of STGCN decreased by 18.19%, 10.45% and 6.24%, respectively compared with the historical average (HA) method, long short-term memory (LSTM) and stacked autoencoders (SAEs). When the forecast window is 2 days, MAE decreased by 9.93%, 3.96% and 4.37%; When the forecast window is 3 days, MAE decreased by 7.02%, 2.47% and 9.20%. The example proves that STGCN can improve the accuracy of delay prediction and provide reference and guidance for airport delay decision.
- flight delay prediction /
- deep learning /
- airport network /
- graph convolutional network /
- gated linear units

HTML全文

图 1 机场网络图结构的网络数据

Figure 1. Network data of airport network graph structure

下载: 全尺寸图片幻灯片

图 2 STGCN结构

Figure 2. Architecture of STGCN

下载: 全尺寸图片幻灯片

图 3 机场网络

Figure 3. Airport network

下载: 全尺寸图片幻灯片

图 4 机场网络邻接矩阵${\boldsymbol{W}}$可视化

Figure 4. Visualization of adjacency matrix ${\boldsymbol{W}}$ in airport network

下载: 全尺寸图片幻灯片

图 5 不同预测模型的预测曲线

Figure 5. Prediction curves of different forecasting models

下载: 全尺寸图片幻灯片

图 6 不同机场的预测结果

Figure 6. Prediction results of different airports

下载: 全尺寸图片幻灯片

图 7 三种延误程度下的各机场预测误差比较

Figure 7. Comparison of forecast errors in airports under three delay levels

下载: 全尺寸图片幻灯片

图 8 不同测试日的预测结果

Figure 8. Predicted results on different test days

下载: 全尺寸图片幻灯片

表 1 机场航班延误数据示例

Table 1. Example of airport flight delay data

数据特征	示例
年	2019
月	1
日	13
机场代码	AUS
机场离港准点率/%	83.83

下载: 导出CSV

表 2 不同预测模型预测结果的评价指标对比

Table 2. Comparison of evaluation indexes for prediction results by different forecasting models

模型	MAE			MAPE/%			RMSE
模型	1 d	2 d	3 d	1 d	2 d	3 d	1 d	2 d	3 d
HA	5.899	5.899	5.899	7.898	7.898	7.898	7.991	7.991	7.991
LSTM	5.389	5.532	5.624	7.105	7.324	7.443	7.387	7.633	7.741
SAEs	5.147	5.556	6.041	6.871	7.317	7.88	7.268	7.572	8.101
STGCN	4.826	5.313	5.485	6.399	7.069	7.307	6.842	7.467	7.689

下载: 导出CSV

表 3 STGCN与SAEs预测结果评价指标比较

Table 3. Comparison of evaluation indexes between STGCN and SAEs

挑选测试日	MAE		MAPE/%		RMSE
挑选测试日	STGCN	SAEs	STGCN	SAEs	STGCN	SAEs
淡季中延误日	5.31	6.33	7.45	8.93	6.33	7.06
淡季低延误日	4.88	6.31	5.55	7.14	6.39	8.24
旺季中延误日	5.22	9.01	8.83	14.83	7.65	11.95
旺季低延误日	3.33	4.72	4.02	5.57	4.62	11.95

下载: 导出CSV

参考文献(23)

[1]	PYRGIOTIS N, MALONE K M, ODONI A. Modelling delay propagation within an airport network[J]. Transportation Research Part C:Emerging Technologies, 2013, 27: 60-75. doi: 10.1016/j.trc.2011.05.017
[2]	DING J L, LI H F. The forecasting model of flight delay based on DMT-GMT model[J]. Physics Procedia, 2012, 33: 395-402. doi: 10.1016/j.phpro.2012.05.080
[3]	HENRIQUES R, FEITEIRA I. Predictive modelling: Flight delays and associated factors, Hartsfield-Jackson Atlanta International Airport[J]. Procedia Computer Science, 2018, 138: 638-645. doi: 10.1016/j.procs.2018.10.085
[4]	吴仁彪, 赵娅倩, 屈景怡, 等. 基于CBAM-CondenseNet的航班延误波及预测模型[J]. 电子与信息学报, 2021, 43(1): 187-195. doi: 10.11999/JEIT190794 WU R B, ZHAO Y Q, QU J Y, et al. Flight delay propagation prediction model based on CBAM-CondenseNet[J]. Journal of Electronics & Information Technology, 2021, 43(1): 187-195(in Chinese). doi: 10.11999/JEIT190794
[5]	CAI Q, ALAM S, DUONG V N. A spatial-temporal network perspective for the propagation dynamics of air traffic delays[J]. Engineering, 2021, 7(4): 452-464. doi: 10.1016/j.eng.2020.05.027
[6]	HAO L, HANSE M, ZHANG Y, et al. New York, New York: Two ways of estimating the delay impact of New York Airports[J]. Transportation Research Part E:Logistics and Transportation Review, 2014, 70: 245-260. doi: 10.1016/j.tre.2014.07.004
[7]	GUO Z, YU B, HAO M Y, et al. A novel hybrid method for flight departure delay prediction using random forest regression and maximal information coefficient[J]. Aerospace Science and Technology. 2021, 116: 106822.
[8]	THIAGARAJAN B, SRINIVASAN L, SHARMA A V, et al. A machine learning approach for prediction of on-time performance of flights[C]//2017 IEEE/AIAA 36th Digital Avionics Systems Conference (DASC). Piscataway: IEEE Press, 2017: 1-6.
[9]	罗谦, 张永辉, 程华, 等. 基于航空信息网络的枢纽机场航班延误预测模型[J]. 系统工程理论与实践, 2014, 34(S1): 143-150. doi: 10.12011/1000-6788(2014)s1-143 LUO Q, ZHANG Y H, CHENG H, et al. Study on flight delay prediction model based on flight networks[J]. Systems Engineering-Theory & Practice, 2014, 34(S1): 143-150(in Chinese). doi: 10.12011/1000-6788(2014)s1-143
[10]	王春政, 胡明华, 杨磊, 等. 基于Agent模型的机场网络延误预测[J]. 航空学报, 2021, 42(7): 452-465. WANG C Z, HU M H, YANG L, et al. Airport network delay prediction based on Agent model[J]. Acta Aeronautica et Astronautica Sinica, 2021, 42(7): 452-465(in Chinese).
[11]	KHANMOHAMMADI S, CHOU C, LEWIS H W, et al. A systems approach for scheduling aircraft landings in JFK airport[C]//2014 IEEE International Conference on Fuzzy Systems (FUZZ-IEEE). Piscataway: IEEE Press, 2014: 1578-1585.
[12]	POLSON N G, SOKOLOV V O. Deep learning for short-term traffic flow prediction[J]. Transportation Research Part C:Emerging Technologies, 2017, 79: 1-17. doi: 10.1016/j.trc.2017.02.024
[13]	GU Y L, LU W Q, QIN L Q, et al. Short-term prediction of lane-level traffic speeds: A fusion deep learning model[J]. Transportation Research Part C:Emerging Technologies, 2019, 106: 1-16. doi: 10.1016/j.trc.2019.07.003
[14]	吴仁彪, 李佳怡, 屈景怡. 基于双通道卷积神经网络的航班延误预测模型[J]. 计算机应用, 2018, 38(7): 2100-2106. doi: 10.11772/j.issn.1001-9081.2018010037 WU R B, LI J Y, QU J Y. Flight delay prediction model based on dual-channel convolutional neural network[J]. Journal of Computer Applications, 2018, 38(7): 2100-2106(in Chinese). doi: 10.11772/j.issn.1001-9081.2018010037
[15]	徐冰冰, 岑科廷, 黄俊杰, 等. 图卷积神经网络综述[J]. 计算机学报, 2020, 43(5): 755-780. doi: 10.11897/SP.J.1016.2020.00755 XU B B, CEN K T, HUANG J J, et al. A survey on graph convolutional neural network[J]. Chinese Journal of Computers, 2020, 43(5): 755-780(in Chinese). doi: 10.11897/SP.J.1016.2020.00755
[16]	LI W, WANG X, ZHANG Y W, et al. Traffic flow prediction over muti-sensor data correlation with graph convolution network[J]. Neurocomputing, 2021, 427: 50-63. doi: 10.1016/j.neucom.2020.11.032
[17]	DENG S J, JIA S Y, CHEN J. Exploring spatial-temporal relations via deep convolutional neural networks for traffic flow prediction with incomplete data[J]. Applied Soft Computing, 2019, 78: 712-721. doi: 10.1016/j.asoc.2018.09.040
[18]	YU B, YIN H T, ZHU Z X. Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting[C]// Proceedings of the Twenty-Seventh International Joint Conference on Artificial Intelligence. Freiburg: International Joint Conferences on Artificial Intelligence Organization, 2018: 3634-3640.
[19]	YIN X Y, WU G Z, WEI J Z, et al. Multi-stage attention spatial-temporal graph networks for traffic prediction[J]. Neurocomputing, 2021, 428: 42-53. doi: 10.1016/j.neucom.2020.11.038
[20]	冯宁, 郭晟楠, 宋超, 等. 面向交通流量预测的多组件时空图卷积网络[J]. 软件学报, 2019, 30(3): 759-769. doi: 10.13328/j.cnki.jos.005697 FENG N, GUO S N, SONG C, et al. Multi-component spatial-temporal graph convolution networks for traffic flow forecasting[J]. Journal of Software, 2019, 30(3): 759-769(in Chinese). doi: 10.13328/j.cnki.jos.005697
[21]	屈景怡, 叶萌, 渠星. 基于区域残差和LSTM网络的机场延误预测模型[J]. 通信学报, 2019, 40(4): 149-159. doi: 10.11959/j.issn.1000-436x.2019091 QU J Y, YE M, QU X. Airport delay prediction model based on regional residual and LSTM network[J]. Journal on Communications, 2019, 40(4): 149-159(in Chinese). doi: 10.11959/j.issn.1000-436x.2019091
[22]	王鑫, 吴际, 刘超, 等. 基于LSTM循环神经网络的故障时间序列预测[J]. 北京航空航天大学学报, 2018, 44(4): 772-784. doi: 10.13700/j.bh.1001-5965.2017.0285 WANG X, WU J, LIU C, et al. Exploring LSTM based recurrent neural network for failure time series prediction[J]. Journal of Beijing University of Aeronautics and Astronautics, 2018, 44(4): 772-784(in Chinese). doi: 10.13700/j.bh.1001-5965.2017.0285
[23]	LV Y S, DUAN Y J, KANG W W, et al. Traffic flow prediction with big data: A deep learning approach[J]. IEEE Transactions on Intelligent Transportation Systems, 2015, 16(2): 865-873.