基于网络搜索信息的多模态数据驱动航空客流集成预测

Multi-modal Data-driven Air Passenger Flow Integrated ForecastingBased on Internet Search Information

为了对机场旅客吞吐量进行更高精度的预测,提出了一种基于网络搜索信息的“分解-重构-集成”组合预测新方法。首先,采用平均影响值和时差相关分析法对机场旅客吞吐量相关的网络搜索关键词进行筛选,合成综合搜索指数。其次,利用改进的自适应白噪声完备集合经验模态分解(ICEEMDAN)方法分别将机场旅客吞吐量和综合搜索指数分解为若干子模态序列,依据子序列的样本熵值重构为高、中、低频序列。以搜索指数中的不同频率成分作为辅助输入信息,分别对机场旅客吞吐量的高频和中频序列采用麻雀搜索算法优化的BP神经网络(SSA-BP)模型进行预测,而低频序列采用自回归分布滞后模型进行预测,最后将不同频率序列预测值用SSA-BP模型进行综合集成得到最终的预测值。通过实证发现,该组合预测新方法能显著提高预测的精度,并表现出较好的鲁棒性。

Airport passenger throughput mainly refers to the number of passengers carried by flights arriving and departing.On the one hand,the airport passenger throughput can directly reflect the size and passenger capacity of the airport.On the other hand,it can reflect the degree of social and economic development of the city and its surrounding areas.However,the internal resources of the airport are relatively limited.Check-in,baggage check tracking,safety checks,waiting point scheduling and emergency response strategies all depend on the space-time distribution of the unanticipated passenger flow.By predicting the airport passenger flow timely and accurately,the management can take a precaution measures,dispatch and arrange the airport resources effectively and reasonably.This way,the airport can save operating costs,reduce the waiting time of passengers queuing,and improve passenger satisfaction.In order to predict the airport passenger throughput with higher accuracy,this paper proposes anew“decomposition-reconstruction-integration”combined forecasting method based on internet search information.First,the mean impact value and time difference correlation analysis methods are employed to screen the internet keywords related to the airport passenger throughput,and the correlation between the search volume of each keyword and the original air passenger flow data is investigated to determine the best lag period,and then a comprehensive search index is constructed.Secondly,the ICEEMDAN data decomposition method is used to decompose the airport passenger throughput and comprehensive search index into several sub-modal sequences,which are reconstructed into high,medium and low frequency sequences according to the sample entropy value of the sub-sequences.Taking the different frequency components of the search index as auxiliary input information,we predict the high frequency and medium frequency sequences of the airport passenger throughput by using the BP neural network model optimized by the sparrow search algorithm(SSA-BP),while the low frequency sequence is predicted with autoregressive distributed lag model,and the ultimate forecasting value is obtained by integrating the predicted values of different frequency components with SSA-BP model.This paper focuses on the monthly passenger throughput of Xi’an Xianyang Airport and Chengdu Shuangliu Airport as the research object,takes the data from January 2011 to December 2019(data from Wind database)as the sample set,and employs the search volume of the Baidu keywords which directly related to the airport or related to the corresponding urban scenic spots as the auxiliary prediction information.Through the empirical research,it is found that:(1)The prediction model based on the“decomposition-reconstruction-integration”framework has obtained smaller values in the three horizontal prediction indicators of RMSE,MAPE and MAE.(2)Compared with the model without Baidu search information,the model with Baidu search information has significantly improved the accuracy of horizontal and directional prediction.(3)Through further comparison of multi-step prediction results,it is found that the internet search information always has well auxiliary prediction ability for the passenger throughput of the two airports.Thus,the new combined forecasting model proposed in this paper can significantly improve the prediction accuracy and is strong in robustness.The prediction results of the model in this paper can provide certain decision-making reference for airport managers,and the proposed prediction method can also be used for short-term prediction of tourist flow in popular tourist attractions.