Machine Learned Atmospheric Force Model Training

This technical report investigates the development of a machine learning (ML) model for atmospheric drag forces affecting objects in Low Earth Orbit (LEO). It utilizes Two-Line Elements (TLEs) data to create a training set based on the orbital decay of various objects due to atmospheric drag. The report outlines the process of training the ML model using historical decay data, emphasizing the importance of accurately modeling atmospheric conditions to improve predictions of orbital trajectories. Regression tests are conducted to compare the performance of the ML model against traditional propagation models. The report also discusses the significance of solar weather in influencing atmospheric density and the challenges posed by geomagnetic storms. Additionally, it describes the structure of TLE data and the role of neural networks in refining atmospheric models. The findings aim to enhance the accuracy of predictions related to collision avoidance, remote sensing, and re-entry scenarios for LEO objects.