Negative Imaginary Output Feedback Control for Attitude Stabilization of Reusable Launch Vehicle With Deep Koopman Predictor

The strong nonlinear coupling, parameter uncertainties, and external disturbances pose substantial challenges in attitude stabilization of reusable launch vehicles (RLV). To solve these problems, a two-stage deep Koopman residual predictor based on a small perturbation linearization model is proposed in this article to obtain a high-precision dynamic model of RLV. The nonlinear cross-channel coupling terms neglected by local linearization are compensated while a linear structure suitable for controller synthesis is retained. Subsequently, a dynamic output feedback controller is designed based on negative imaginary (NI) theory calculated with linear matrix inequality. The effects of parameter uncertainties and external disturbances are fully considered. Numerical simulation results demonstrate that the modified Koopman-based model achieves much higher modeling accuracy than the traditional small perturbation linearization method. The proposed NI-based control scheme can quickly converge the states of RLV to the equilibrium position and effectively suppress the influence of disturbances. It provides a reliable control solution for RLVs' attitude stability.