Maximising runway capacity by mid-term prediction of runway configuration and aircraft sequencing using machine learning

Maximising runway capacity is one of the essential measures to meet the growing traffic demand. Runway capacity maximisation is an open challenge in the literature due to a complex and non-linear interplay of many factors which are stochastic in nature such as wind, weather, and arrival and departure sequence of aircraft. However, an effective sequencing of arrivals and departures may condense the service time on runway, thereby generating opportunities for new landing or take-off slots, which may maximise the runway capacity. In addition, sequencing of arrivals and departures optimised for a predicted runway configuration, given the weather and wind conditions, may lead to maximising the runway capacity. First, I develop an optimisation method, using aircraft position shifting and path-planning, for aircraft sequencing for a single runway airport. The proposed method can provide an optimal aircraft sequence, for both arrivals and departures, such that it minimises the total inter-arrival and departure times and, consequently, maximises the runway throughput. The proposed method implements several arrival/departure sequencing strategies, i.e., constraint position-shifting with one, two and N-positions, and First Come First Serve (FCFS) in order to obtain an optimal sequence (i.e., a sequence with the lowest process time). The novelty of the sequencing model is to incorporate the Standard Terminal Arrival Routes (STAR) for path planning and sequencing of arriving aircraft at Final Approach Fix (FAF) and departing aircraft sequence at the runway threshold. The simulation result demonstrates that the model can increase up to 15% of the runway capacity compared with the commonly used aircraft sequencing technique (i.e., FCFS). Second, a runway configuration (i.e., a set of runways active in a specific period in a multi-runaway system) plays a vital role in determining runway capacity. Thus, I developed an evolutionary computation (Cooperative Co-evolutionary with Genetic Algorithm) algorithm for determining which runway configuration is most suitable for processing a given optimal aircraft sequence (arrival-departure), such that the runway capacity is maximised in a multi-runway system. The proposed algorithm models and uses Runway Configuration Capacity Envelopes (RCCEs) which defines arrival and departure throughputs. RCCE helps in identifying the unique capacity constraints based on which runways are used, for example, arrivals/departures or both. In the proposed evolutionary algorithm (CCoGA), the runway configuration and aircraft sequence are modelled as two species which interacts and evolve co-operatively to yield the best populations (combination) for maximising runway capacity. The fitness function for the optimal sequence species is to reduce the total process time for a given runway configuration, while fitness function for the runway configuration species is to maximises the total capacity for the given optimal sequence. The simulation results show that CCoGA can provide trade-off solutions with multiple runway configurations, for a given arrival-departure sequence, which can lead to capacity maximisation. Third, the weather conditions at an airport play a major role in determining the runway configuration which then has a significant impact on its runway capacity. Typically, aircraft take-off and landing operations use the runways which are most closely aligned with the wind direction, speed and other factors (e.g., cloud ceiling, visibility). However, selecting a runway configuration is a challenging task because it requires not only wind/weather (current and predicted) conditions but also the arrival/departure sequence (active and anticipated) at an airport. Predicting a suitable runway configuration under the operating conditions and a given traffic distribution may be useful for maximising the runway capacity. To achieve that, I first propose a Machine Learning (ML) model for predicting a suitable runway configuration given wind/weather and arrival/departure sequence. The simulation results demonstrate the accuracy of the prediction (98%) and usefulness of the ML techniques for assisting Air Traffic Controllers (ATCs) to choose certain runway configuration based on real-world weather data. In the final part of this thesis, I extend the ML model for forecasting runway configuration and develop a capacity estimation model for estimating associated capacity in a medium-term horizon (6 hours). The proposed model incorporates influencing factors (i.e., wind, visibility, cloud ceiling, and operation time) as well as all possible runway layouts for predicting the most suitable configuration. As a case study for Amsterdam Schiphol Airport, one month of airport weather data and associated runway configurations are processed to train and test the developed ML model. The results, on two days of sample traffic data, demonstrate the prediction accuracy of the ML model of up to 98%. Also, it is demonstrated that the ML predicted configuration can accommodate an additional number of flights (i.e., up to 20 flights) within one hour. This shows the viability and benefits of using ML approach for maximising runway capacity.

Grantor dc:publisher

UNSW, Sydney

Year dc:date

2018

Author dc:creator

• Ahmed, Md Shohel

• Machine learning
• Runway capacity maximisation
• Runway configuration prediction
• Aircraft sequence
• Evolutionary computation
• Terminal airspace path-planning

Statement dc:rights

• open access
• CC BY-NC-ND 3.0
• free_to_read

Licence

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by-nc-nd/3.0/au/

Language dc:language

EN

Identifier

https://doi.org/10.26190/unsworks/20928

OAI identifier oai:identifier

oai:unsworks.library.unsw.edu.au:1959.4/61221