Stochastic modeling of physical drag coefficient – Its impact on orbit prediction and space traffic management

Ambitious satellite constellation projects by commercial entities and the ease of access to space in recent times have led a dramatic proliferation low-Earth traffic. It jeopardizes safety long-term sustainability, necessitating better domain awareness (SDA). Correct modeling uncertainties force models orbital states, among other things, is an essential part SDA. For objects orbit (LEO) region, uncertainty dynamics mainly emanate from limited knowledge atmospheric drag-related parameters variables. In this paper, which extends work Paul et al. (2021), we develop feed-forward deep neural network model for prediction drag coefficient full range attitude (i.e., pitch ∈ (-90°,+90°) yaw (0°,+360°)). The simultaneously predicts mean standard deviation well-calibrated. We use numerically simulated physical data training our network. numerical simulations are carried out using test particle Monte Carlo method diffuse reflection with incomplete accommodation gas-surface interaction model. Modeling well-known CHAllenging Minisatellite Payload (CHAMP) satellite. Finally, approach propagate CHAMP over three-day period under various scenarios investigate distribution radial, along-track, cross-track errors caused uncertainty. key takeaways paper - (a) constant cannot be used reliable SDA purposes, (b) stochastic machine learning allow computation coefficients timely manner while providing estimates.