Current studentsExisting staffContact us
 

Predicting aircraft landing time in extended-TMA using machine learning methods

Simple item page
dc.contributor.authorDhief, I.
dc.contributor.authorWang, Z.
dc.contributor.authorLiang, M.
dc.contributor.authorAlam, S.
dc.contributor.authorSchultz, M.
dc.contributor.authorDelahaye, D.
dc.contributor.conference9th International Conference for Research in Air Transportation (ICRAT) 2020 (15 Sep 2020 : Virtual)
dc.date.issued2020
dc.descriptionLink to a related website: https://hal-enac.archives-ouvertes.fr/hal-02907597/file/ICRAT_paper.pdf, Open Access via Unpaywall
dc.description.abstractAccurate prediction of aircraft arrival times is one of the fundamental elements for air traffic controllers to manage anoptimal arrival and departure sequencing on the runway, reduce flight delays, and achieve a good collaboration with airports andairlines. In this work, we analyze the feature engineering problem to predict Aircraft Landing Time (LDT) in Extended- TMAwith machine learning models. Two main contributions are highlighted in this work. First, the impact of different features in LDTprediction is investigated. Second, a machine learning prediction model is presented to predict LDT. Our case of study is the ETMA of Singapore Changi Airport (WSSS) with a radius of 100NM. Firstly, data analysis is conducted to check the availabilityof different resource data, as well as cleaning the raw trajectory data. Then, feature construction and extraction are discussed indetails, machine learning prediction models are proposed to address the LDT prediction. The experimental results show that 4sets of features play a significant impact on LDT prediction for primary runway-in-use, they are: (1) Control intent: trafficdemand, current traffic density, and adjacent flow; (2)Weather: surface wind; (3) Trajectory: the position of aircraft; (4)Seasonality: parts of a day and a week. Moreover, comparing three Machine Learning algorithms, in our study case, Extra- Treesis the best prediction algorithm compared with other machine learning models in terms of Root Mean Square Error (RMSE) andMean Absolute Error (MAE). It is also found that Machine learning models perform much better than the current operationalsystem. In summary, two main conclusions are drawn from the present work. First, predicting the aircraft LDT is stronglycorrelated with the TMA density at the flight operation time. Second, feature selection with domain knowledge and expertopinions is very important, and with good features, the model is less sensitive to the choice of machine learning algorithm.
dc.identifier.citation9th International Conference for Research in Air Transportation (ICRAT), 2020, pp.1-8
dc.identifier.urihttps://hdl.handle.net/11541.2/143102
dc.language.isoen
dc.publisherICRAT
dc.publisher.placeUS
dc.relation.fundingCivil Aviation Authority of Singapore , Program number: Aviation Transformation Program
dc.rightsCopyright 2020 ICRAT
dc.source.urihttps://hal-enac.archives-ouvertes.fr/hal-02907597/file/ICRAT_paper.pdf
dc.subjectTerminal Maneuvering Area
dc.subjecttrajectory prediction
dc.subjectmachine learning
dc.subjectdata mining
dc.subjectaircraft landing time
dc.titlePredicting aircraft landing time in extended-TMA using machine learning methods
dc.typeConference item
pubs.publication-statusPublished
ror.mmsid9916418501301831

Files

Collections