Adaptive Neural Network And Fault Tolerant Control Of Nonlinear Systems

With the development of science and technology,the model of control system is also becoming more and more complex.The original linear system theory is not enough to describe the physical model,and the corresponding control theory is facing new challenges.In practice,we are often faced with complex nonlinear systems,which are characterized by model nonlinearity,time-varying and uncertain terms.Adaptive control is a good choice for systems with nonlinear models,and a reasonable adaptive term is designed to achieve stable control objectives when the model parameters are unknown.For nonlinear systems with unknown structure,neural network function approximation theory makes it an effective approximation tool for unknown nonlinear model.When the system has large structure,we use the idea of distributed control to divide the system to multi-subsystems,then we use neural network controller to approximate the unknown model in each subsystem and the whole system can be stable.The research of adaptive neural network is aimed at the control research of unknown nonlinear system.The adaptive neural network which combines the approximation theory of neural network and the online learning ability of adaptive control can realize the control problem of complex system.Therefore,how to design a reasonable adaptive neural network controller to realize the stability,reliability and applicability of the system has become a intresting problem for further discussion and research.In this paper,theory of nonlinear system,adaptive control,neural network,and distributed event triggered control method are used to study the global stability of adaptive neural network,event driven control strategy and distributed neural network control.The main work is as follows:(1)For the control of MIMO(Multi-input Multi-Output)high order nonlinear system,we assume that the approximation error of neral network control less than a function related to the system state,then design RBF neural network on the relevant items for unknown nonlinear approximation;neural network weight value and approximation error are handled by adaptive laws;the controller for high order system is designed based on Backstepping technology;a global Lyapunov function is given with fault tolerant control to ensure that the system can keep stable in a designed adaptive controller.(2)In view of the large amount of computation and communication in the adaptive neural network and the Backstepping process,we design the corresponding event trigger control strategy for the actual network control.In this part,we design static and dynamic event driven control strategies based on adaptive neural networks.In the static strategy,A control structure is designed based on the given event-triggered constant,to ensure the system can save communicaiton resource and unlimited shock problem does not occur in the control of the error;in the dynamic strategy,dynamic trigger parameter is given to ensure the stability of the system and makes the number of events triggered relatively stable when the system control input changes rapidly.(3)For large scale systems,a single neural network structure is not suitable for controlling the whole large-scale system.Based on the idea of distributed control,the whole system is divided into several subsystems,and then the corresponding adaptive neural network control algorithm is designed.Considering that only part of the subsystem may have an executor in a distributed system,the system need to be reorganized.In view of the connection disturbance and the unknown model in each subsystem,a unified neural network structure is designed to approach it.Finally,we design a distributed control algorithm for the high speed train system by combining the corresponding control algorithm with the related problems in the high speed train system.