Agent-based simulation for aircraft stand operations to predict ground time using machine learning

Punctual and reliable aircraft ground handling operations at the airports significantly contribute to efficient traffic flows in the air traffic network. Any improved prediction of aircraft ground times can help to reduce local delays and delay propagation in the network by taking into account the forecast of future operational states for adjusted planning and delay mitigation strategies. In our work, we target to predict aircraft ground times at their stands by machine learning algorithms, where the complete turnaround sub-processes and domain knowledge are input for the models. We develop two types of models, the first type is regression-oriented that intends to forecast the exact aircraft ground time. And the second one is classification-oriented, which attempts to confirm aircraft off-block time adherence. An agent-based approach is applied to generate some synthetic data, besides, we also obtain an actual aircraft ground handling dataset from a certain European airport to validate our models. Finally, the interpretable method for the machine learning models is used to analyse the feature importances, and the feature affections on the prediction results. The results show that our classification model is capable to predict accurate aircraft off-block time adherence.