Managing Storm Uncertainties for a Safe and Efficient Air Transport System: Storm-Avoidance Aircraft Trajectory Optimization. (PID2021-122323OB-C31)
Universidad Rey Juan Carlos (Madrid, Spain)
ABOUT
Within the general objective of the coordinated project, the objective of this subproject is to optimize the storm-avoidance aircraft trajectories predicted by Subproject 2 taking into account the storm evolution predicted by Subproject 3. The problem will be solved in the 3D space, i.e., the optimized storm-avoidance trajectories will have three spatial components plus time. The storm-avoidance trajectory optimization method developed in this subproject is based on optimal control and, besides the weather forecast, it considers multiple aircraft, described by precise dynamical models, the operational constraints, and the uncertainties of the problem. It additionally allows a probabilistic upper bound on the risk of encountering a cell of a storm to be included in the formulation and is endowed with decision-making and learning capabilities.
Decision making processes will be introduced in the storm-avoidance aircraft trajectory optimization process to establish for instance which, among two or more actions, should be taken to solve a contingency and are usually modelled with logical constraints in disjunctive form. This type of constraints also arise in modelling storm-avoidance, temporal or spatial separation among aircraft and operational constraints. Standard modeling techniques tackle logical constraints in disjunctive form using binary or integer variables in the formulation of the optimal control problem. Solving the resulting mixed-integer optimal control problem is computationally very expensive due to its combinatorial complexity. To overcome this difficulty, in this subproject, the logical constraints in disjunctive form will be transformed into inequality and equality constraints that involve only continuous auxiliary variables, which greatly reduces the computational cost of finding a solution.
Aircraft trajectory optimization methods require the definition of an objective functional and a set of constraints in order to achieve the desired behavior of the aircraft in performing an action. Specifying such an objective functional for a complex manoeuvre, such as avoiding the cells of a storm, which are actually moving and shape-changing obstacles to avoid, is a difficult task. Additionally, since they rely on nonlinear programming, they can get stuck in local optima and consequently they can fail in finding the optimal trajectory. To overcome these difficulties, the storm-avoidance trajectories predicted in Subproject 2 are included in the aircraft trajectory optimization process. This is done by combining aircraft trajectory optimization with imitation learning, also called learning from demonstration. In this method, introduced for robot programming, the robot learns a behavior to perform an action implicitly from the trajectories demonstrated by human experts. With this combination, the storm-avoidance trajectory optimization is less dependent on the choice of the objective functional and the set of constraints and thus offers an alternative way to achieve the desired behavior.
RESEARCH TEAM
Ernesto Staffetti Giammaria
ernesto.staffetti@urjc.es
Alberto Olivares González
alberto.olivares@urjc.es
WORK TEAM
Almudena Buelta Méndez
almudenajose.buelta@urjc.es
María Cerezo Magaña
maria.cerezo@urjc.es
Marius Marinescu
marius.marinescu@urjc.es
Jaime de la Mota Sanchis
jaime.delamota@urjc.es
EXPECTED SCIENTIFIC IMPACT
The objective of this subproject is to design a method for solving the problem of optimising the aircraft trajectories predicted by Subproject 2 by Subproject 2 capable of avoiding storms taking into account the evolution of storms based on the evolution of storms based on the forecasts provided by Subproject 3. Subproject 3. The method for optimisation of aircraft trajectories capable of storm avoidance to be capable of storm avoidance that will be developed in this sub-project will be equipped with learning and decision-making capabilities, which are of particular importance. which are of particular importance.
In the presence of a storm, several decisions must be made in a short time interval. short interval of time. As a result of the disruption of as a result of the disruption of flight plans, conflicts may arise and the airspace available for conflict resolution is limited. for conflict resolution is limited. The sequencing order sequencing order of aircraft in an airway may be altered due to the manoeuvres required to avoid thunderstorms and the necessary to avoid thunderstorms, and the presence of a thunderstorm may prevent the may prevent operational restrictions from being met. In these situations, a trajectory In these situations, an aircraft trajectory optimisation system with decision-making capability would be of great benefit to the end-users: pilots, flight operators and pilots, flight operators and air traffic controllers.
This subproject will implement a specific numerical technique capable of finding solutions in computational times compatible with tactical applications. tactical applications.
Ideal storm cell avoidance manoeuvres cannot be easily described by a flight procedure, nor can they be easily determined using trajectory optimisation techniques, but they can be learned from demonstrations which, in this context, consist of storm avoidance trajectories performed by other aircraft in similar scenarios. This sub-project will follow an approach that combines trajectory optimisation techniques based on optimal control with imitation learning techniques. The rationale behind this transition from purely optimal control-based techniques to a mixed approach is threefold. Firstly, this approach allows the experience of pilots avoiding storm cells to be taken into account in the trajectory optimisation process. Secondly, it has been shown that imitation learning is an effective mechanism for reducing computational time in finding optimal solutions to trajectory optimisation problems.
In fact, observed trajectories are feasible solutions that can reduce the probability of obtaining local optima. Thirdly, by combining optimisation-based methods with imitation learning in trajectory optimisation, trajectory generation will rely less on optimal control techniques, which are based on the specification of an objective functional and constraints, making them more accessible to end-users with limited experience in this field.
A user-friendly decision support system for the optimisation of storm-avoidance aircraft trajectories would facilitate the application of this technology in air traffic management in order to reduce the consequences that convective weather events can have on the air traffic system, ensuring the efficient and safe use of airspace.
Storm-ATS 