Abstract: To address the highly nonlinear and uncertain characteristics in the models of hypersonic flight vehicles, an intelligent con-trol method based on probabilistic inference for learning is proposed. Without relying on precise models, this method utilizes Gaussian Process to learn dynamic characteristics from flight data, constructing a probabilistic dynamics model that reflects uncertainties to achieve long-term predictions of state sequences. Considering the cascaded flight control architecture, a hier-archical optimization mechanism for the inner and outer loop control strategies is designed, and the analytical gradients for parameter optimization are derived, which reduces the parameter search dimensionality while ensuring the closed-loop stabil-ity of the system. For different operating tasks across a wide flight envelope, an online adaptive controller is constructed by fitting the offline-optimized control parameter sets. Numerical simulations show that the proposed method achieves high learning efficiency, and the learned results retain command tracking capability under aerodynamic parameter perturbations.

Key words: Keywords: Hypersonic Vehicles, Probabilistic Inference for Learning, Intelligent Control, Gaussian Process, Hierarchical Optimization