Research On Adaptive Control Method Of Near Space Vehicle Based On Neural Network

The control problem of near space vehicles is a hot issue in the aerospace field.Because of its strong coupling,large disturbances,nonlinearity and fast time-varying characteristics,its controller design is also a difficult problem.In this thesis,different controllers based on neural network is designed for the attitude control of near-space vehicles,considering system uncertainties and nonlinear constraints.First,the mathematical model of the aircraft is established.Considering the flying environment of the aircraft,the basic mathematical model of the near-space vehicle is established from the aspects of kinematics and dynamics.Then the uncertainty of the model was analyzed and the mathematical model of the aircraft with uncertainty was established.Considering the uncertainty of the system,a dynamic surface controller based on neural network is designed.The uncertainty of the system has been simplified.It is assumed that the uncertainty of the control matrix and external disturbances can be compensated by the neural network.The stability is proved by stability analysis,and the control effect is verified by simulation.Then,this thesis makes an in-depth study on the actual engineering requirements in the attitude control of the aircraft,and considers that there are nonlinear constraints on the part of the system and the control inputs.At the same time,it also considers that the uncertainties of the system cannot be completely compensated by the neural network,and designed an adaptive law to deal with the uncompensable part of the neural network.In the case of considering nonlinear constraints,the neural network will learn the error information due to the nonlinear constraints that do not need to learn.The auxiliary system is used to adjust the dynamic error of each step to eliminate the influence of the nonlinear constraints on the neural network.The final stability analysis and simulation prove that all signals of the system are semi-globally consistent and ultimately bounded.Finally,the combination of active disturbance rejection control and neural network is used to deal with the attitude control of the aircraft.The neural network is used to replace the expansion state in the expansion state observer,and the network structure is used to identify the disturbance of the system instead of a single value,which ensures that the recognition result is smoother.On the premise of ensuring system performance,the smoothness of control commands is improved.Combined with the neural network,it can directly carry out the three-channel coupling design,which overcomes the shortcomings of the traditional active disturbance rejection control that can only be designed with a single variable.Finally,the stability of the observer is analyzed using Lyapunov stability theory and the stability of the closed-loop system is verified by numerical simulation.