Journal of Aerospace Operations

Motivated by challenges in strategic traffic flow management, a stochastic network model is introduced for the spatiotemporal evolution of weather impact at a strategic time horizon. Specifically, a model that represents weather-impact propagation using local probabilistic influences is shown to capture the rich dynamics and inherent variability in weather impact at the spatial and temporal resolution of interest. This model serves as a simulation tool to generate stochastic weather-impact trajectories for strategic air traffic management. To develop the simulator, first the underlying influence model concept is introduced. Next, an approach for parameterizing the model from probabilistic weather forecast data is developed, such that statistics of generated weather/weather-impact trajectories match the forecasts at snapshot times. An example weather-impact simulator for a particular bad-weather event, namely a long-duration convective weather event in Atlanta Center on 26 September 2010, is also developed. The framework shows promise for e.g. 1) managing weather contingencies for the airspace over a full day; 2) adjusting forecast resolution in data-limited areas for special management requirements; and 3) evaluating variability in airspace performance.

Fast-time simulation Traffic Flow Management (TFM) models can play a significant role in helping to identify best practices for weather impact scenarios, to apply these practices, and to review their efficacy. A model that can enforce Traffic Management Initiatives (TMIs) to provide sufficient realism is highly desirable for these activities. The model must also be comprehensively weather-aware so as to faithfully reflect the effect of convective and non-convective weather on air traffic. We have leveraged the ability of the Dynamic Airspace Routing Tool (DART), a superfast-time simulation model, to objectively evaluate the potential benefits (or added impacts) of alternative TFM strategies in support of operationally relevant what-if testing possibilities. We describe a methodology enabled by these simulation capabilities to objectively, and with great detail, evaluate adverse-weather-related TMI utility and potential alternatives with the aim of building a library of ‘best practices’. This overall approach, developed for visual, operator-driven Strategic TFM analysis through comparison of baseline and alternative scenarios, can also be augmented by a parametric search, i.e. partial optimization, of feasible TMI solutions for still greater testing and automation support for impact management decision-making. We discuss how this simulation model has already been effectively applied to support alternative-response analyses for air traffic operations. The opportunities for Strategic TFM to evolve to a more consistent and efficient operation under this alternative support paradigm are discussed.

Ground delay programs (GDPs) are often initiated in the U.S. National Airspace System to balance demand with capacity at a capacity-constrained arrival airport. A GDP assigns departure delay at origin airports to modulate demand at the destination airport. Usually, one GDP affects hundreds of flights. The substantial impact of GDPs on flight operations leads to our research interest in predicting GDP initiation. In this paper, we identify variables that play a significant role on GDP initiation decisions and quantify their impact using logistic regression. We consider lead times of from 1 to 4 hours and specify a logistic model for each lead time. This allows us to provide a GDP initiation prediction for flight operators for up to a 4-hour time horizon. Further, using cross-validation, we compare the predictions of these models, including a weighted accuracy, true positive rate, and precision. We find that the GDP initiation predictions over the longer time horizon are only slightly less reliable than that of one hour into the future. Whatever the time horizon, however, the model predictions are often incorrect, either predicting a GDP when one is not implemented or vice versa.

We use historical data to build two types of stochastic model of hourly Ground Delay Program (GDP) implementation for the three main airports near New York City: Newark Liberty International, LaGuardia, and John F. Kennedy International. The models predict the probability that a GDP will be initialized or canceled in a given hour based on hundreds of features describing the situation at the airport, including features describing forecasted weather conditions and scheduled traffic. One is a regularized logistic regression model that ignores system dynamics and the other model is based on inverse reinforcement learning, so it considers system dynamics and the impact that GDP implementation actions have over time on some metrics. We evaluate the models based on two objectives: their ability to predict and simulate GDP implementation decisions in historical test data sets. As is expected based on the motivation for and objective of each type of model, regularized logistic regression models make superior predictions while simulations controlled by inverse reinforcement learning models produce average metric values that more closely match those produced by historical GDP implementation decisions. Finally, we draw insights about GDP implementation from the trained model parameters. For example, parameters of both models suggest while weather conditions and values of performance metrics such as airborne delay play a role in GDP decision making, parameters related to the predictability or continuity of existing GDP plans are also important.

Air Traffic Flow Management (ATFM) aims at structuring traffic in order to reduce congestion in airspace. Congestion being linked to aircraft located at the same position at the same time, ATFM organizes traffic in the spatial dimension (e.g. route network) and in the time dimension (e.g. sequencing and merging of aircraft flows taking off or landing at airports).

The objective of this paper is to develop a methodology that allows the traffic to self-organize in the time and space dimensions when demand is high. This structure disappears when the demand diminishes. In order to reach this goal, a multi-agent system has been developed, in which aircraft cooperate to structure traffic. Multi-agent systems have several advantages, including a good resilience when confronted with disruptive events.

In this system, three algorithms have been implemented, aiming at reducing traffic complexity in three different ways. The first algorithm allows aircraft agents flying on a route network to regulate speed in order to reduce the number of conflicts, a conflict occurring when two aircraft do not respect separation norms. The second algorithm allows aircraft to solve conflicts when the traffic is not structured by a route network. The third algorithm creates temporary local route networks allowing to structure traffic.

The three algorithms implemented in this multi-agent system allow to decrease overall traffic complexity, which becomes easier to manage by air traffic controllers. This algorithm was applied on realistic examples and was able to structure traffic in a resilient way.