基于时空长短期记忆网络的机场航班延误预测 李娟,曾维理*,刘丹丹,羊钊,丛玮2 a.南京航空航天大学民航学院,江苏南京211106:b.飞友科技有限公司,江苏南京211106) 瘸要:准确预测杋场航班延误情况,对于全面协调空管、机场、航空公司的运行 至关重要。目前,长短期记忆(LSTM)神经网络在各领域得到广泛应用,特别是 在预测方面,精度明显高于传统的机器学习预测方法。本文将深度学习方法进行推 广和应用到机场航班延误预测,提岀一种时空LSTM神经网络航班延误预测方 通过考虑关联杋场在时间和空间的相关性,将所有关联杋场的到达延误和岀发延误 作为特征变量,能充分从高维特征变量中捕获延误传播内在杋理,从而提高延误预 测的精度。以美国2015年至2018年主要机场延误数据作为实验数据,实验结果表 明,在不同时长预测精度上,本文方法在预测精度上均优于目前主流延误预测方法。 关键词 延误预测;深度学习;LSIM网络;时空变量 Spatial-Temporal Long Short-term Memory Networks for Airport Flight Delay Prediction Li Juan, Zeng Wei-li, Liu Dan-dan, Yang Zhao, Cong Wei (1. College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, Nanjing, 211106: 2. Variflight Technology Company Abstract: Accurately predicting airport flight delays is important for overall comprehensive coordination of airports, airlines and air traffic management. However, due to the complexity of the impact on flight delays, the existing mainstream methods have lower prediction accuracy Inspired by the method of traffic flow prediction based on deep learning, a method based on deep lstM for flight delay prediction is proposed. Different from the existing mainstream methods, this paper firstly applies the Spatial-Temporal Long Short-term Memory Networks (LSTM) to flight delay prediction, and considers the correlation of time and space of associated airports, and features the arrival delay and departure delay of all associated airports. The algorithm can totally obtain information from high-dimensional data Based on the delay data of the major airports in the United States from 2015 to 2018 as experimental data. Experimental Results demonstrated that the airport delay prediction method is superior to the current mainstream delay prediction method in predicting accuracy in different time length prediction key words: delay prediction; deep learning, LSTM Network; spatial-temporal variable 基金项目:中央高校基本科研业务费资助(NoNS2018044)和国家自然科学基金(No.51608268) 通讯作者:曾维理( zwlnuaaanuaaedu cn)基于时空长短期记忆网络的机场航班延误预测 李娟,曾维理*,刘丹丹,羊钊，丛玮 2 （a. 南京航空航天大学民航学院，江苏 南京 211106； b. 飞友科技有限公司，江苏 南京 211106） 摘 要： 准确预测机场航班延误情况，对于全面协调空管、机场、航空公司的运行 至关重要。目前，长短期记忆（LSTM）神经网络在各领域得到广泛应用，特别是 在预测方面，精度明显高于传统的机器学习预测方法。本文将深度学习方法进行推 广和应用到机场航班延误预测，提出一种时空 LSTM 神经网络航班延误预测方法。 通过考虑关联机场在时间和空间的相关性，将所有关联机场的到达延误和出发延误 作为特征变量，能充分从高维特征变量中捕获延误传播内在机理，从而提高延误预 测的精度。以美国 2015 年至 2018 年主要机场延误数据作为实验数据，实验结果表 明，在不同时长预测精度上，本文方法在预测精度上均优于目前主流延误预测方法。 关键词： 延误预测; 深度学习; LSTM 网络；时空变量 Spatial-Temporal Long Short-term Memory Networks for Airport Flight Delay Prediction Li Juan, Zeng Wei-li, Liu Dan-dan,Yang Zhao1 ,Cong Wei2 (1. College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, Nanjing, 211106；2. Variflight Technology Company） Abstract: Accurately predicting airport flight delays is important for overall comprehensive coordination of airports, airlines and air traffic management. However, due to the complexity of the impact on flight delays, the existing mainstream methods have lower prediction accuracy. Inspired by the method of traffic flow prediction based on deep learning, a method based on deep LSTM for flight delay prediction is proposed. Different from the existing mainstream methods, this paper firstly applies the Spatial-Temporal Long Short-term Memory Networks (LSTM) to flight delay prediction, and considers the correlation of time and space of associated airports, and features the arrival delay and departure delay of all associated airports .The algorithm can totally obtain information from high-dimensional data. Based on the delay data of the major airports in the United States from 2015 to 2018 as experimental data. Experimental Results demonstrated that the airport delay prediction method is superior to the current mainstream delay prediction method in predicting accuracy in different time length prediction. key words: delay prediction; deep learning; LSTM Network; spatial-temporal variable 基金项目：中央高校基本科研业务费资助（No.NS2018044）和国家自然科学基金（No.51608268） 通讯作者：曾维理(zwlnuaa@nuaa.edu.cn)