Projecte llegit
Títol: Predicción de la hora de aterrizaje y tiempo de taxi-in en el Aeropuerto de Barcelona-El Prat mediante técnicas de machine learning
Estudiants que han llegit aquest projecte:
CANTARONE RUIZ, ROBERTO (data lectura: 09-09-2026)- Cerca aquest projecte a Bibliotècnica
CANTARONE RUIZ, ROBERTO (data lectura: 09-09-2026)Director/a: ERITJA OLIVELLA, ANTONI-JOSEP
Departament: FIS
Títol: Predicción de la hora de aterrizaje y tiempo de taxi-in en el Aeropuerto de Barcelona-El Prat mediante técnicas de machine learning
Data inici oferta: 13-01-2026 Data finalització oferta: 13-09-2026
Estudis d'assignació del projecte:
- GR ENG SIS TELECOMUN
- GR ENG SIST AEROESP
- GR ENG TELEMÀTICA
| Tipus: Individual | |
| Lloc de realització: Fora UPC | |
| Supervisor/a extern: Marta Sánchez Cidoncha | |
| Institució/Empresa: CRIDA | |
| Titulació del Director/a: Enginyer/a Aeronàutic/a | |
| Paraules clau: | |
| Machine Learning, Airport Operations, Data Analysis, Regression Models | |
| Descripció del contingut i pla d'activitats: | |
| Efficient airport operations are essential for reducing delays, fuel consumption, and emissions. One component of these operations is the taxi-in time, defined as the time between aircraft landing and arrival at the assigned gate. This duration can vary significantly due to runway configuration, traffic density, airport layout, aircraft characteristics, and operational procedures.
In this bachelor's thesis, you will develop a machine learning model to predict aircraft taxi-in and approach times at Barcelona-El Prat Airport using historical flight and airport operational data. You will analyse operational factors influencing taxi-in performance, touch-down times, and approach delays, and translate them into features usable by machine-learning algorithms. The work combines knowledge of aviation operations with data-driven modelling, providing insight into how advanced analytics can support airport decision-making. |
|
| Overview (resum en anglès): | |
| This Bachelor's Thesis develops a machine learning-based tool to improve aircraft arrival prediction at Josep Tarradellas Barcelona-El Prat Airport (LEBL). The methodology is based on two sequential models: the first estimates the remaining time from the aircraft entry into the Extended Arrival Sequencing and Metering Area (E-ASMA) until landing, while the second predicts the subsequent taxi-in time to the parking stand.
The models are developed using historical trajectory data, operational information and meteorological variables from LEBL operations. After data cleaning, merging, feature engineering and encoding, XGBoost regression models are trained. Their performance is evaluated through global metrics, disaggregated analyses under different operational conditions and a SHAP (SHapley Additive exPlanations) analysis to identify the most influential variables. The results are also compared with models generated using H2O AutoML. Model 1 achieves a mean absolute error (MAE) of 1.50 min and reduces by 45.0% the error of the available estimated landing time (ELDT). Its accuracy decreases mainly under high E-ASMA occupancy conditions, while differences between airlines remain limited. Model 2 achieves an MAE of 47.63 s and shows a stronger dependence on runway, wake turbulence category and airline, which are indirectly related to the ground movement. The sequential architecture does not show significant error propagation between both models. H2O AutoML provides slightly better results, although the differences remain limited, making XGBoost competitive while offering greater interpretability. Overall, the results show that updating the prediction at E-ASMA entry significantly improves the landing-time estimate, while taxi-in prediction remains mainly limited by the available information on the aircraft ground movement. |
|