ResNet-Enhanced TD3-Based Coordinated Control and Transformer-Based Online Fault Detection System

Rocket attitude and trajectory control has long been a key topic in the field of control systems due to the harsh working environments of rockets and the stringent demands for control responsiveness and robustness. These challenges make stabilizing rocket attitude and trajectory particularly difficult. This paper proposes a multi-thruster coordinated control method for rocket attitude and trajectory based on the TD3 architecture, incorporating ResNet into the reinforcement learning framework, achieving superior results compared to traditional models. Furthermore, to enhance system robustness and provide real-time status updates for operators, the paper introduces an online fault detection system based on a transformer network. This system quantitatively assesses the operational status of each thruster in a short time, offering additional insights for system control and enabling operators to monitor the system’s status more accurately and promptly.