Cislunar satellite motion prediction via hybrid parametric and deep learning models

Abstract

The cislunar region, increasingly recognized for its strategic role for scientific, commercial, and military purposes, has emerged as the focus of numerous space missions planned over the next decade. The spacecraft motion in this region is governed by the complex interaction of gravitational forces by the Earth, the Moon, and the Sun, as well as perturbations such as solar radiation pressure and gravitational interactions with other celestial bodies. An accurate assessment of the dynamics of stable and unstable orbital regions and the prediction of absolute and relative satellite-to-satellite motion are essential for the success of missions and operational safety. This work introduces a novel hybrid approach that integrates a parametric analysis based on deep learning techniques to isolate regions in which the states of the satellites do not diverge and predict their evolution as a function of time. By defining non-dimensional parametric surfaces and curves in the Earth-Moon rotating reference frame, the method isolates orbital configurations that lead to non-divergent trajectories. The analysis enabled the identification of a region of initial conditions to generate high-fidelity trajectories. This set is then used to train a deep learning model capable of efficiently predicting both absolute and relative satellite-to-satellite states. The proposed approach significantly enhances Space Domain Awareness capabilities, addressing the challenges of managing an increasing number of cislunar missions and mitigating risks such as orbital collisions and instability.