Machine learning for global trajectory optimisation
Can we use machine learning to extract knowledge and information from the process of trajectory optimisation? If so, can we use this knowledge for the design of new trajectories, or to enhance and improve the quality of the optimisation process?
Study Description
This project investigates the use of machine learning tools for the global optimisation of interplanetary trajectories. In particular, we study how approximation models can help to speed up and improve the results of trajectory optimisation processes and how they can help us extract useful information about the input/output relationships.
First, we use objective function landscape approximation methods to study how we can speed up and improve the results of the optimisation of spacecraft interplanetary trajectory problems. These multimodal problems, recently introduced in the global optimisation community, can be very complex and characterised by the prevalence of many local optima and, in the worst cases, a heavy computation load involved with the calculation of the objective value of a given input vector. We use artificial neural networks trained online using input vector / objective function value pairings generated by the optimisation process as a substitute to part of the calls to the original objective value. First results confirm that the resulting hybrid algorithm does not degrade the quality of the optimisation results while offering a computational speed-up.
Second, we use data analysis and machine learning techniques to automatically extract interesting correlations among orbital characteristics (e.g., time of flight, semi-major axis, inclination, etc.) from a large sample size of optimal trajectories. Machine learning can offer automatic classification that can prove meaningful for human mission designers.