Séminaire McTAO : Bernd Dachwald (FH Aachen University) – 8 septembre 2023

Optimization of Deep Space Missions Using Neural Networks

Bernd Dachwald (FH Aachen University)

Vendredi 8 septembre, 14h00, salle Coriolis (Galois).

Abstract. Among the main goals for the optimization of deep space missions is the minimization of flight-time and propellant mass. Innovative deep space missions use ion thrusters, which need much less propellant than chemical thrusters, or, in the future, solar sails, which need no propellant at all. Searching optimal deep space trajectories for such spacecraft with continuous thrust is a difficult and time-consuming task. The solution quality of traditional optimal-control methods depends strongly on an initial guess. For complex trajectories, an adequate initial guess is hard to find. Even if the optimizer converges, the solution is typically close to the initial guess and far from the global optimum.
A neural network in combination with an evolutionary algorithm can be successfully applied for finding near-globally optimal trajectories for continuous-thrust spacecraft. Such evolutionary neurocontrollers attack the problem from the perspective of artificial intelligence and machine learning. They do neither require an initial guess nor the involvement of an expert in astrodynamics and optimal control. Evolutionary neurocontrollers search the solution space of all possible trajectories much deeper than a human expert (using conventional optimizers) can do and thus find complex trajectories that are closer to the global optimum.
In the talk, the principle of the method is explained and its performance is demonstrated for various deep space missions.