Improving Climb Performance Prediction in Air Traffic Control with Machine Learning and Full Flight Simulator Verification

A deep feedforward network has been used to predict the flight level with look ahead time up to six minutes for climbing flights. Representing features were developed from operational Mode S Enhanced Surveillance data. In parallel, using certified A340 and A319 Full Flight Simulators, complete climb profiles from Take Off to Top of Climb have been recorded. Certain parameters were modified, e.g. Take Off weight or head- and tailwind conditions. The features from the Full Flight Simulator were fed to the neural network which has been trained with the operational data. The predictions of the flight level in both cases are compared. Because the parameters in the simulator were known and controlled, it allows us to infer the settings of the operational traffic and to improve the features by adding wind information and recognizing the acceleration phase.