A hybrid intelligent attitude control framework for spacecraft enhanced by generalized neural network prediction

Jupiter exploration represents a strategic frontier in deep space missions. Its orbit is subject to strong perturbations and rapid temporal variations within an unknown space environment, posing severe challenges to the precision and reliability of spacecraft attitude control. Traditional model-free adaptive control relies on error feedback, inherently suffering from time delay. Existing neural network adaptive schemes tightly couple intelligent models with control laws and stability proofs, making them difficult to adapt to the rapid iteration pace of AI technologies and lacking theoretical guidelines for universal safety integration. To address this, this paper proposes a universal neural network prediction-enhanced hybrid intelligent attitude control framework for spacecraft (applicable to all prediction networks). Through a decoupled dual-channel design, an independent neural network feedforward channel achieves active prediction compensation for unknown disturbances, overcoming the latency bottleneck of traditional methods. A model reference adaptive feedback channel ensures the global stability of the closed-loop system. Based on Lyapunov theory, this paper rigorously proves the system’s globally consistent ultimate bounded stability. Crucially, this stability conclusion is independent of the prediction model’s internal structure, requiring only bounded prediction error to achieve “plug-and-play” model adaptability. Numerical simulations demonstrate that this framework significantly enhances system transient performance under fast-transient, strong disturbances, establishing a new technological paradigm for high-precision, high-reliability attitude control of spacecraft in Jupiter’s extreme environments.