Attitude control method for launch vehicle operation based on reinforcement learning

In order to achieve the synergy between periodic self-operation control and task-driven non-periodic random control， an intelligent attitude control method is proposed for launch vehicles， which is based on reinforcement learning model predictive control （RMPC）. Firstly， a break point identification method based on time-frequency analysis is established to address the issue of switching requirements between periodic orbit maintenance and non-periodic maneuver adjustment during rocket flight. By monitoring the spectral characteristics of control command sequences in real-time， the transition boundary between periodic and non-periodic time processes is accurately located， which can serve as decision-making basis for control mode switching. Secondly， a RMPC framework integrating adaptive attention mechanism is proposed， which can dynamically adjust the prediction model structure and iterative optimization steps based on fuzzy neural network according to the type of control task， while ensuring prediction accuracy and controlling model complexity. Finally， an environment aware time-domain decision-making method is designed to enhance the adaptability of RMPC in complex environments by adaptively adjusting the predicted and control time domain lengths through online evaluation of external interference strength and system state uncertainty. The method is applied to closed-loop control simulation of typical flight scenarios of launch vehicles， and the experimental results show that trajectory tracking control precision can be maintained by using the proposed method in both periodic orbit maintenance phase and non-periodic maneuver phase.