Data-driven uncertainty quantification and propagation for probabilistic trajectory planning

Kurzfassung

One of the main objectives of Trajectory-Based Operations (TBO) is to increase the predictability of the aircraft behavior within the Air Traffic Management (ATM) system. However, most systems involved in TBO (such as flight planning systems) focus on proposing deterministic trajectories in the strategic phase, not taking into account the uncertainty factors that affect the trajectory prediction process in the tactical phase. Consequently, there is an increased frequency of updates and modifications to trajectories in later planning phases, which leads to degraded stability, resulting in an overall decrease of the performance of the ATM network. In this presentation, a data-driven methodology will be introduced for characterizing the uncertainties affecting the development of an aircraft trajectory, together with their integration into a stochastic trajectory predictor for obtaining robust sets of probabilistic trajectories from an initial flight plan. Additionally, this methodology employs data assimilation models that capture updated information from the air traffic system to reduce the present uncertainty. First, the main sources of uncertainty for aircraft trajectories will be identified and quantified using historical flight instances for a full year of pan-European air traffic. After quantifying these sources of uncertainty, it will be possible to evaluate the potential variations for a flight plan given the probability distributions for uncertain factors affecting the flight. Instead of applying computationally demanding methods, such as Monte Carlo simulations, for calculating all possible trajectories, a stochastic trajectory predictor is proposed that makes use of the characterization of trajectory uncertainty to compute probabilistic trajectories given an initial flight plan. The stochastic trajectory predictor uses arbitrary Polynomial Chaos Expansion (PCE) theory and the point collocation method to find polynomials describing the aircraft trajectory for the initial flight plan as a function of the identified uncertain factors. Therefore, the quantified uncertainty sources can be fitted in the polynomials to find a reduced set of probabilistic trajectories that are robust and resilient to potential variations in the tactical phase. Complementing this, a set of advanced data-assimilation models based on machine learning techniques are integrated to provide accurate estimations for some of the uncertain factors based on the last available status of the air traffic system. These estimates reduce the uncertainty spectrum for important variables in the trajectory prediction process and help adapting the resulting probabilistic trajectories to the current system status. Finally, a study case is introduced in which the proposed methodology is implemented. This study includes the results of analyzing the probabilistic trajectories for one city-pair and supports the idea of integrating probabilistic trajectories as a key enabler for envisioned TBO concepts and modern airline operations planning