Abstract:

Aiming at the control problem of the flexible hypersonic vehicle with constrained aerodynamic surfaces, an intelligent control scheme based on neural adaptation is proposed. In the design process of the velocity subsystem, the deep reinforcement learning algorithm is used to adjust the proportional integral derivative (PID) parameters online, and the intelligent PID control strategy is given to reduce the dependence on the model parameters. For the altitude subsystem, the neural adaptive method is used to approximate the unknown functions and uncertain terms of the model considering the dynamic characteristics of the aerodynamic surfaces. To deal with the constraint problem of aerodynamic surfaces, a large number of sample data sets are generated using nonlinear model predictive control as an optimal allocation template, and a deep neural network obtained through offline training is employed to replace the process of solving complex optimization problems and control allocation. In addition, by introducing an adaptive super-twist differentiator to handle external disturbances, the robustness of the system is enhanced. The stability of the controller is proved by using the Lyapunov method. The simulated results show that the proposed method can quickly calculate the control commands and realize high-precision tracking control.

Key words: hypersonic vehicle, neural adaptation, intelligent control, deep reinforcement learning, deep neural network (DNN)