融合空间关系与时间序列特征的民航旅客量预测算法  

FORECASTING CIVIL AVIATION PASSENGER BY COMBINING SPATIAL RELATION AND TIME SERIES FEATURE

作　　者：吴丽娜[1,2] 冯迪 李忠虎[1,2] Wu Lina;Feng Di;Li Zhonghu(Travelsky Technology Limited,Beijing 101318,China;Key Laboratory of Intelligent Passenger Service of Civil Aviation,CAAC,Beijing 101318,China)

机构地区：[1]中国民航信息网络股份有限公司,北京101318 [2]民航旅客服务智能化应用技术重点实验室,北京101318

出　　处：《计算机应用与软件》2022年第12期298-302,317,共6页Computer Applications and Software

基　　金：民航科技重大专项(MHRD20160109)。

摘　　要：针对民航航线网络中各机场的空间关系带来的相互影响以及民航旅客量的时间序列特征,建立了一个融合空间关系与时间序列特征的民航旅客量预测算法,采用深度卷积神经网络捕捉民航旅客量的时序和空间信息,采用三次指数平滑模型捕捉民航旅客量时间序列的季节等周期性信息、长期趋势信息及时序信息,并加入日期特征和天气特征作为外部因素,采用全连接网络对上述两个模型的结果及外部因素进行融合。将提出的模型应用于30个典型的民航城市构成的航线网络的未来14天的旅客量预测中,其有效地提高了民航旅客量预测的准确性。To solve the affect of spatial relation of airports in civil aviation airline network and temporal feature of civil aviation passengers, this paper proposes an algorithm for forecasting civil aviation passenger by combining spatial relation and time series feature. The deep convolutional neural network was applied to capture the temporal and spatial information of civil aviation passenger. The triple exponential smoothing model was used to capture the seasonal and other periodic information, long-term trend information and time sequence information of the time series of civil aviation passenger. The date characteristics and weather characteristics were added as external factors, and the results of the above two models and external factors were fused by using a fully connected network. The proposed algorithm was applied to the forecast of civil aviation passengers in the next 14 days of civil aviation composed of 30 typical civil aviation cities. The result shows that it effectively improves the accuracy of civil aviation passenger prediction.

关 键 词：民航 智能交通 旅客量预测 卷积神经网络 三次指数平滑 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术] TP18