Abstract This paper proposes artificial neural network models to predict the arrival/departure capacity of airports. Multilayer perceptron (MLP), recurrent neural networks (RNN), and long short-term memory (LSTM) models have been trained using capacity and meteorological data from Hartsfield–Jackson Atlanta International Airport (ATL) from 2013 to 2017. The models’ predictive performances were validated against the observed capacity of ATL in 2018. The qualitative and quantitative analysis of the trained models confirmed that the artificial neural networks approach is effective in predicting airport capacity. In addition, the transferability of the models for Boston Logan International Airport (BOS) is examined. Capacity prediction performance for BOS measures the transferability of the models trained with the ATL data. MLP showed good transferability without taking any other measures, and RNN and LSTM were able to predict the BOS capacity well after fine-tuning.
Highlights This research investigates air traffic management using artificial neural networks. MLP, RNN, and LSTM are used to predict the airport’s departure/arrival capacity. The models are trained using historical capacity and weather data from ATL. The analysis confirms that the models are effective in airport capacity prediction. With transfer learning, the models can be easily applied to a new airport.
Artificial neural network models for airport capacity prediction
2021-09-11
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
FAA Airport Capacity and Delay Models
NTIS | 1977
|Airport traffic flow prediction method based on graph neural network
Europäisches Patentamt | 2023
|TIBKAT | 1963
|