Spacecraft conjunction assessment optimization using deep learning algorithms applied to conjunction data messages (CDMs)

The lack of global regulations on space debris management during the early days of the space era until the last few decades of the 20th century resulted in a consistent increase in space debris. Spacecraft collisions in orbit and the industry's growing interest in launching constellations of satellites are now exacerbating the problem. To address those concerns, multiple space organisations worldwide have implemented Situational Space Awareness programmes with integrated Conjunction Assessment systems that allow the detection of spacecraft conjunctions with an estimated collision risk probability. While this approach has proved effective in the last two decades, the foreseen increment of artificial space objects in orbit in the coming years will put any existing system under severe stress if the technology does not evolve to match the new demands. The objective of this research is two-fold: it evaluates different architectures used in the field of Deep Learning to increase the accuracy of on-orbit Conjunction Events forecasting. It provides a multi-purpose modular, Machine Learning based Python library to support Conjunction Assessment activities. The results of this study show that simpler cell architectures used in the Recurrent Neural Networks outperform the corresponding Vanilla versions in terms of accuracy for the problem at hand. It also demonstrates that the attention mechanism provides the best performance with up to 40% more accuracy.