ATM network modelling, uncertainty propagation with thunderstorm disruptions

Kurzfassung

In this work, as a part of START, we have developed an ATM network macro-model, allowing us to model the propagation of flight trajectory uncertainties and further assess the impact of disruptive events, i.e., thunderstorms. We utilized data-driven analytics models mimicking the dynamics of epidemic spreading, which is analogous to delay or uncertainty propagation over transport networks. The connections between the operational aspects of the air traffic flow management and the developed meta-model are given as the airports' traffic densities correlated with the infection rates among the individuals; and the capability to absorb the uncertainties of the airports associated with recovery rates. Uncertainties over individual flight trajectories, which are the functions of flight times, have been defined through probabilistic distributions where superposed on the arrival times. Deep learning models have been integrated to capture the nonlinear relationship between the recovery rates, uncertainty accumulation, and disruptive events' attributes. The model allowed us to simulate and analyze the behavior of the network under uncertainty accumulations coming from trajectory uncertainty. Finally, we have used Reinforcement Learning to explore the best actions to enhance the network resiliency, defined through stability theory. From the operational perspective, resiliency is associated with the managing balance between the intervention rate (depending on "the time for washing away the effect of the transition period) and costs. The problem, at this point, transformed into an optimization-based control problem to guarantee convergence over time, meaning the effect of disruptive events dies out eventually. Quick recovery is typically preferred, but it applies significant intervention measures impacting many flights in this case. RL provided us with pinpointing the OD pairs, and the flights require regulatory action such as flight cancelation and aircraft grounding. The case studies are analyzed for the selected time windows chosen in the interval of 1-10 June 2018, where thunderstorms affected large areas of North-West Europe with intense local convective activities.