AbstractAir traffic management (ATM) under its current paradigm is reaching its structural limits considering the continuously growing demand. The need for a decrease in traffic workload opens numerous problems for optimization, from capacity balancing to conflict solving, using many different degrees of freedom, such as re-routing, flight-level changes, or ground-holding schemes. These problems are usually of a large dimension (there are 30 000 daily flights in Europe in the year 2012) and highly combinatorial, hence challenging for current problem solving technologies. We give brief tutorials on ATM and constraint programming (CP), and survey the literature on deploying CP technology for modelling and solving combinatorial problems that occur in an ATM context.

As acknowledged by the SESAR (Single European Sky ATM (Air Traffic Management) Research) program, current Air Traffic Control (ATC) systems must be drastically improved to accommodate the predicted traffic growth in Europe. In this context, the Episode 3 project aims at assessing the performance of new ATM concepts, like 4D-trajectory planning and strategic deconfliction.One of the bottlenecks impeding ATC performances is the hourly capacity constraints defined on each en-route ATC sector to limit the rate of aircraft. Previous works were mainly focused on optimizing the current ground holding slot allocation process devised to satisfy these constraints. We propose to estimate the cost of directly solving all conflicts in the upper airspace with ground holding, provided that aircraft were able to follow their trajectories accurately.We present a Constraint Programming model of this large-scale combinatorial optimization problem and the results obtained with the FaCiLe (Functional Constraint Library). We study the effect of uncertainties on the departure time and estimate the cost of improving the robustness of our solutions with the Complete Air Traffic Simulator (CATS). Encouraging results were obtained without uncertainty but the costs of robust solutions are prohibitive. Our approach may however be improved, for example, with a prior flight level allocation and the dynamic resolution of remaining conflicts with one of CATS' modules.

The en route conflict resolution problem has been modeled in many different ways, generally depending on the tools proposed to solve it. For instance, with purely analytic mathematical solvers, models tend to be very restrictive to respect the inherent limitations of the technology. This paper introduces a new framework that separates the model from the solver so as to be able to: first, enhance the model with as many refinements as necessary to comply with operational constraints; second, compare different resolution methods on the same data, which is a crucial aspect of scientific research.To this aim, our framework generates a benchmark of conflict resolution problems built with various scenarios involving different numbers of aircraft, levels of uncertainties and numbers of maneuvers. We then compare two different optimization paradigms, Evolutionary Algorithm and Constraint Programming, which can efficiently solve difficult instances in near real time, to illustrate the usefulness of our approach.into account all these uncertainties to choose the best trajectories in terms, first, of safety and then, efficiency. These certainties probably explain why the short-term traffic resolution system still relies on human expertise and is not yet automated.Much research has been done on conflict detection and resolution and many papers present models that are so impractical that they strengthen the readers' beliefs that automating the conflict detection and resolution task is unrealistic in the near-term. For example, the approach using repulsive forces described in [Zeghal, 1993] or the B-spline approximation model of [Delahaye et al., 2010] are very interesting on a mathematical level but could hardly be implemented in an operational context. They suppose continuous heading changes, which Flight Management Systems (FMS) are unable to exploit, and do not take uncertainties into account. Pallottino's approach [Pallottino et al., 2002] using mixed integer linear programming (as [Vela et al., 2009, Alonso-Ayuso et al., 2011, Rey et al., 2012) relies on constant speed trajectories that are changed all at once. None of these approaches could deal with realistic trajectory models able to handle evolutive aircraft or trajectory uncertainties.Other approaches like [Durand et al., 1996, Granger et al., 2001 propose to solve conflicts using Evolutionary Algorithms, relying on more realistic models built upon the Base of Aircraft Data (BADA) developed and maintained by EUROCONTROL. These models introduce uncertainties on aircraft speed, climb and descent rate, thus the solver needs to compute many alternative trajectories in real time. Nevertheless, the solver is quite efficient as it can handle complete days of traffic in the European airspace. These algorithms, however, are difficult to compare with other methods because the conflict detection is embedded in the solver. This problem also occurs in Erzberger's approach [Erzberger, 1997], where most of the expertise is focused on the trajectory and maneuver model. Once more, th...

International audienceAir traffic in Europe is predicted to increase considerably over the next decades. In this context, we present a study of the interactions between the costs due to ground-holding regulations and the costs due to en-route air traffic control. We describe a traffic simulator that considers the regulation delays, aircraft trajectories, and air conflict resolution. Through intensive simulations based on traffic forecasts extrapolated from French traffic data for 2012, we compute the regulation delays and avoidance maneuvers according to two scenarios: the current regulations and no regulations. The resulting cost analysis highlights the exponential growth in regulation costs that can be expected if the procedures and the airspace capacity do not change. Compared to the delay costs, the costs of the air traffic control are negligible with or without regulation. The analysis reveals the heavy controller workloads when there are no regulations, suggesting the need for regulations that are appropriate for large traffic volumes and an improved ATC system. These observations motivate the design of a third scenario that computes the sector capacities to find a compromise between the increase in the delay costs due to ground-holding regulations and the increase in the controller workload. The results reveal the sensitivity of the delay costs to the sector capacity; this information will be useful for further research into ATM sector capacity and ATC automated tool design. Finally, because of the growing interest in the free flight paradigm, we also perform a traffic and cost analysis for aircraft following direct routes. The results obtained highlight the fuel and time savings and the spatial restrictions that companies use to avoid congested areas

One of the key challenges towards more automation in Air Traffic Control is the resolution of en-route conflicts. In this article we present a generic framework for the conflict resolution problem that clearly separates the trajectory and conflict models from the resolution. It is able to handle any kind of maneuver and detection models, though we propose our own realistic 3D maneuvers and conflict detection that takes into account uncertainties on the positioning of aircraft. Based on these models, realistic scenarios are built, for which potential conflicts are detected using an efficient GPU-based algorithm. The resulting instances of the conflict resolution problem are provided to the community as a public benchmark.To efficiently solve this problem, we also introduce a generic framework for the cooperation of optimization algorithms. The framework benefits from the various optimization algorithms plugged to it by sharing relevant information among them, and is implemented as a distributed application for better performance. We illustrate its behavior on the conflict resolution problem with the cooperation between a Memetic Algorithm and an Integer Linear Program which consistently outperforms previous approaches by orders of magnitude. Instances with up to 60 aircraft are optimally solved within a few minutes using this framework, while each algorithm taken individually only provides sub-optimal solutions. This cooperative approach thus seems appropriate for application in a real-time context.