Sebastian Gruber, Paul Feichtenschlager, Christoph Georg Schütz,
"Using Genetic Algorithms for Privacy-Preserving Optimization of Multi-Objective Assignment Problems in Time-Critical Settings: An Application in Air Traffic Flow Management"
: Proceedings of the ACM Genetic and Evolutionary Computation Conference 2024 (GECCO 2024), Melbourne, Australia, July 14-18, 2024, ACM Press, Seite(n) 1246 - 1254, 7-2024
Original Titel:
Using Genetic Algorithms for Privacy-Preserving Optimization of Multi-Objective Assignment Problems in Time-Critical Settings: An Application in Air Traffic Flow Management
Sprache des Titels:
Englisch
Original Buchtitel:
Proceedings of the ACM Genetic and Evolutionary Computation Conference 2024 (GECCO 2024), Melbourne, Australia, July 14-18, 2024
Original Kurzfassung:
In air traffic flow management (ATFM), temporarily reduced capacity in the European air traffic network leads to the Network Manager imposing a regulation, meaning that flights are assigned new arrival times on a first-planned, first-served basis. Some flights, however, are more important for airlines and the airport than others due to various reasons, e.g., different numbers of affected passengers across flights. Therefore, optimization of the assignment of flights to available arrival times based on airline and airport preferences has the potential to considerably improve overall efficiency. In the ATFM setting, with its multiple, often competing stakeholders, the inputs for the optimization, e.g., costs of delay, are sensitive information, which must be protected. Furthermore, solutions must be found within the available time frame, which for the flight prioritization problem in ATFM is only in the order of minutes. The privacy-preserving implementation of multi-objective optimization algorithms has considerable computational overhead, which may lead to the optimization not finishing within the deadline. To alleviate this problem, we propose the separation of the search for solutions and the evaluation of the solutions, with only the evaluation requiring a privacy-preserving implementation. Our experimental results suggest good convergence under limited time while protecting sensitive inputs.
Forschungs-