关键词: 交通运输工程；机器学习；数据挖掘；天气因素；航班延误；机场容量

Abstract: For the network structure of civil aviation transportation, evaluating the operation capacity of a single airport can only be used for flight scheduling, while predicting the spatiotemporal distribution of airport network capacity has guiding significance for flight operation control. In order to reflect the impact of complex weather on the airport real-time capacity, the big data of weather and flight operations of China civil aviation was adopted, and the data mining technology was used to extract the real-time capacity of the airport. And the nonlinear mapping relationship between weather and capacity was also studied. A model that converted weather spatiotemporal distribution prediction into airport capacity spatiotemporal distribution prediction was established by using various machine learning algorithms such as neural networks. Moreover, on the basis of analyzing the prediction results of a single model, the neural network weighted feature fusion layer was used to construct a deep ensemble prediction model, which deeply integrated multiple algorithms to further improve the accuracy of real-time airport capacity prediction. Finally, the multi-seasonal weather data of airport network in Eastern China was used to verify the accuracy of the proposed model in predicting the real-time capacity of the airport under complex weather conditions. It is found by the research that the prediction accuracy of each model decreases with the increase of unit time, and the deep neural network has a strong ability to learn nonlinear weather characteristics. The construction of deep neural network through fusion ensemble learning has effectively improved the prediction accuracy of real-time airport capacity. The proposed deep ensemble prediction model satisfies the requirements of obtaining the real-time capacity of future airports in a short time, which can provide decision-making basis for civil aviation transportation and reduce flight delays.It satisfies the requirement of obtaining the real-time capacity of future airports in a short time, and can provide decision basis for the dynamic operation of the airport and reduce flight delays.

Key words: traffic and transportation engineering; machine learning; data mining; weather factors; flight delay; airport capacity