Please use this identifier to cite or link to this item:
|Title:||Prediction of runway configuration change transition timings using machine learning approach||Authors:||Lau, Max En Cheng||Keywords:||Engineering::Aeronautical engineering::Aviation||Issue Date:||2021||Publisher:||Nanyang Technological University||Source:||Lau, M. E. C. (2021). Prediction of runway configuration change transition timings using machine learning approach. Final Year Project (FYP), Nanyang Technological University, Singapore. https://hdl.handle.net/10356/150292||Abstract:||With the increasing focus on maximising airport capacity to meet the exponential growth in demand for aviation services, runway capacity optimisation has become a key area to focus on. For existing operations, optimal runway configuration is a direct way to maximise capacity. To achieve this, the transition time during a runway configuration change must be considered to determine if a change can better achieve effective traffic management in terminal airspace. While recent studies have proposed novel models to determine the ideal runway configurations, the gap in existing research of transition timings still limits part of the decision making process when optimising for maximum overall throughput, including accounting for disruptions during the changeover process. Hence an improved understanding of transition time is needed to better improve and predict runway configuration changes. In this report, a comprehensive machine learning model was produced in order to predict the transition times required for a change in runway configuration, taking into account the dynamic weather and traffic conditions. This involved data consolidation and feature engineering from multiple sources to compile meteorological, positional and arrival data for flights arriving into Philadelphia International Airport from December 2019 to January 2020. An intermediate clustering using Gaussian Mixture Model predicted runway assignments, achieving a 96.9% accuracy compared with manual assignments and 89.9% validity with operational runways. For the machine learning prediction, 3 types of ensemble methods, Random Forest Regressor, AdaBoost and Gradient Tree Boosting, were utilised to achieve prediction R2 scores of at least 0.8 out of a possible 1, across 7 selected runway configuration changes. For these 7 changes, AdaBoost and Gradient Tree Boosting each provided 1 best performing ensemble, with Random Forest Regressor achieving the best scores for the remaining 5 runway configuration change with a maximum of R2 score of 0.949. Random Forest Regressor was also the most consistent method, delivering R2 scores of above 0.8 for all 7 runways regardless. A combination of wind direction, wind speed and height of the lowest cloud base factors were the best in terms of predictive power of the model, with a maximum combined feature importance of 0.892. All 7 configuration changes regarded a combination of some cloud and wind factors in the top 3 most important features. The Gradient Tree Boosting had a relatively higher accuracy compared to the other ensemble methods for the runway configuration change with the least number of changes to be made. AdaBoost generally outperformed the rest when these runway configuration changes were more complex.||URI:||https://hdl.handle.net/10356/150292||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Student Reports (FYP/IA/PA/PI)|
Updated on Jan 20, 2022
Updated on Jan 20, 2022
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.