 |
 |
This thesis work presents the design and implementation of traditional and interpretable machine learning (ML) models for the future European Space Agency LISA interferometer and, more in general, for space missions lacking in-situ observations of interplanetary parameters. The developed models are built upon the lessons learned with the precursor mission of LISA, named LISA Pathfinder, and exploit the correlation between galactic cosmic-ray (GCR) flux measurements and fluctuations observed in the solar wind speed and interplanetary magnetic field intensity.
The ultimate result presented in this thesis work is the estimation via interpretable ML tools of the solar wind speed for LISA based on preceding observations of GCR flux variations and interplanetary magnetic field intensity. All the enabling activities carried out to build the interpretable model are also detailed here, including the development of novel explainable artificial intelligence (XAI) algorithms to extract knowledge from opaque ML regressors and of a Python framework to support XAI. The design of preliminary ML models dedicated to GCR predictions is also reported. All the developed models are trained with the data provided by the LISA Pathfinder and ACE missions, to demonstrate the feasibility of obtaining interplanetary parameter reconstructions based on the joint observations gathered by different satellites.
The outputs provided by the presented interpretable regressor, expressed as linear equations describing the solar wind speed in terms of GCR flux variations and interplanetary magnetic field intensity, constitute the first attempt to identify a quantitative relationship between these interplanetary parameters.
reference thesis
