Reinforcement Learning-Based Self-Driven Model Predictive State Constraint Control for Spacecraft Deorbit

In addressing the multi-constraint optimization control problem of active orbital decay for spacecraft, this study proposes a Model Predictive Control (MPC) approach that incorporates reinforcement learning, employing a self-triggered and adaptive prediction horizon. To tackle coupled orbital state constraints, this paper introduces a method based on reinforcement learning utility reward functions. Furthermore, this method incorporates the driving force into the prediction step and control mechanism of Model Predictive Control, forming a self-driven and adaptive step-length Model Predictive Control approach. At each triggering moment, the controller solves the spacecraft’s optimal orbital decay control problem (OCP), providing an optimal control sequence and determining the next triggering time and prediction horizon. This approach enhances the efficiency of solving Model Predictive Control, simultaneously improving its capabilities in handling multi-constraint predictive control problems. Simulation results ultimately confirm the effectiveness of the proposed algorithm.