This paper presents two approaches to real-time System Identification (SysID), applied to challenging space use-cases to enable complex disturbance rejection. The first approach consists of a novel Sparse Identification of Nonlinear Dynamics algorithm based on Nonlinear AutoRegressive eXogenous models (NARX-SINDy), using an L1-regularised regression to identify the model structure while accounting for real-time, discretization, and observability constraints. The second approach, used as baseline, consists of an Extended Kalman Filter
(EKF) assuming narrowband disturbance spectral content. To evaluate these approaches, two industrial use-cases are considered. The first involves nonlinear propellant sloshing during orbit control manoeuvres, and the second involves solar array drive mechanism (SADM) behaviour in the presence of backlash and nonlinear stiffness effects. In both cases, the disturbance spectral content is close to the controller bandwidth, where it is not possible to increase the controller gain without compromising performance and robustness. Performance is assessed using real ight data from Airbus’ eet and representative simulation campaigns. The NARX-SINDy method outperforms the EKF when the disturbance deviates from the narrowband approximation, or when significant time-varying effects are present, and it enables active disturbance rejection on both use-cases. Implementation feasibility is verified using a processor in the loop test based on a space-grade LEON4 processor.