Modeling Longitudinal Evolution of Decommissioned Geostationary Satellites using Neural Networks

Abstract

This study uses neural networks to explore the intricate longitudinal progression of decommissioned geostationary satellites. The goal is to model and predict satellites' longitudinal dynamics across time dimensions. Historical satellite longitude data undergoes thorough preprocessing to train time series neural networks in both single-input and 3-input configurations for all six decommissioned satellites, yielding comprehensive longitudinal behavior insights. Results reveal impressive outcomes: average Mean Squared Error (MSE) between predicted and measured longitudes is 1.55x10-3, with regression close to unity. This convergence implies a strong alignment between the neural network methodology employed and the intricate problem domain. These results accentuate the suitability and effectiveness of the chosen neural network approach in addressing the challenges posed by decommissioned geostationary satellite trajectory modeling. The study's implications span various fields. Insight into long-term orbital shifts aids in understanding satellite behaviors, enhancing trajectory predictions and decision-making in satellite management and space technology advancement. Additionally the research emphasizes the importance of accurate predictions about satellite behavior after decommissioning. This contributes to better mission planning, resource optimization, and more efficient strategies for dealing with space debris.