As the airspace is getting increasingly crowded worldwide, the capacity management of airports is more important than ever. To avoid unnecessary and costly delays, it is crucial for airports to have well timed strategies and reliable arrival predictions. To achieve this, airports need accurate long-term trajectory predictions such that arrival times can be estimated with high precision. Countless factors such as weather conditions, restricted fly areas and air traffic control clearances cause route uncertainties, making it difficult to predict long-term trajectories accurately. In this thesis, a bidirectional LSTM recurrent neural network is proposed to solve a sequence-to-sequence learning problem and predict the most likely route flown by the aircraft before take off. The network is trained on historical flight plans only, making it easy to implement. The data exists of incoming flights on Schiphol international airport within Europe. Different Hyperparameters are tested to improve performance of the network.
Graduated: July 2022