Benchmarking Different Global Optimisation Techniques For Preliminary Space Trajectory Design
- Paper number
IAC-07-C1.3.01
- Author
Dr. Dario Izzo, European Space Agency (ESA)/ESTEC, The Netherlands
- Coauthor
Dr. Tamas Vinko, Hungary
- Coauthor
Mr. Claudio Bombardelli, University of Padova, Italy
- Coauthor
Dr. Claudio Bombardelli, Advanced Concepts Team, The Netherlands
- Year
2007
- Abstract
A number of global optimisation techniques has been recently proposed to approach trajectory design problems such as multiple gravity assists, deep space manoeuvres, low thrust transfers and so on. Approaches based on heuristics such as Ant Colony Optimisation, Particle Swarm Optimisation (PSO), Differential Evolution (DE), Genetic Algorithms (GA), Simulated Annealing (SA) but also deterministic solvers have been proposed for different trajectory problem instances. This paper is an effort to introduce standard benchmark problems suitable to study the performances of global optimisation algorithms in spacecraft trajectory optimisation. We introduce the multiple gravity assits (MGA) problem with powered swing-bys, the MGA problem with deep space manoeuvres and two particular instances of the low-thrust trajectory optimisation problem. We describe the exact implementation of these problems into a black-box function returning the objective function f and the constraints g for a given decision vector x. The problem can then be seen as:
min: f ( x) subject to: g ( x )≤ 0 where x ∈ Ω ⊂ R. The dimensions of the problem clearly depend on the number of swing-bys considered and on the trajectory model used. We select a number of problem instances including real cases such as Cassini, Rosetta, Messenger, but also academic cases such as an asteroid grand tour, an asteroid deflection mission, an Earth-Mars cargo mission, an interstellar mission. We then test on each one of these problems standard implementations of global optimisation solvers. In our comparison we include PSO, multiple PSO, DE, GA and SA. Carefully selecting the number of objective function evaluations allowed and the tuneable parameters for each algorithm we find that some implementations are outperforming others in a wide range of different problems instances. .1cm In order to make our test problems and algorithms available to research groups we share on-line all the code of the benchmark problems and of the solvers used to produce these results.
- Abstract document
- Manuscript document
IAC-07-C1.3.01.pdf (🔒 authorized access only).
To get the manuscript, please contact IAF Secretariat.