Real-time Prediction of Aircraft Boarding

Reliable and predictable ground operations are essential for 4D aircraft trajectories. The ground trajectory of an aircraft primarily consists of the handling processes at the stand, defined as the aircraft turnaround. The turnaround is mainly controlled by operational experts, but the critical aircraft boarding is driven by the passengers' experience and willingness or ability to follow the proposed procedures. Using a complexity metric for the evaluation of the boarding progress we developed a machine learning approach to predict the boarding progress over time. A calibrated boarding model provides reliable input data for our recurrent neural network approach for both learning and prediction. The input contains of discrete time series, which inherently considers the dynamic passenger behavior. In particular we use a Long Short-Term Memory network to learn this dynamical behavior over time with regards to the given boarding progress indicators.