Model-based Event Triggered Deterministic Learning Control For A Class Of Discrete Time Strict Feedback Nonlinear Systems

With the development of network informatization,more and more resources are transmitted through the network.The traditional time-triggered control periodically transmits data at each discrete moment,and this process transmits a large amount of redundant information.To effectively reduce the transmission of redundant information,the event-triggered control proposed in recent years has attracted wide attention because of its feature of non-periodic transmission of necessary signals.For some practical systems with nonlinearity,uncertainty,and strong coupling,the current research on event-triggered control focuses on the zero-order hold mechanism that keeps the transmission signal unchanged during the event-triggering interval.Under the premise of a certain understanding of the system structure,the control performance caused by this mechanism still has some disadvantages,and it is worth doing more in-depth research.For a class of systems with limited network resources,this thesis proposes model-based event-triggered control scheme.The breakthrough of this scheme lies in the design of the trigger mechanism and the trigger model.The zero-order hold mechanism is to keep the signal transmitted from the other side of the network unchanged;and the model-based mechanism is to construct a model which is coordinated with the original system structure to provide dynamic signals.Based on the above analysis,the work of this thesis is summarized as follows:First of all,in the network environment,this thesis respectively designs a zero-order hold and model-based event-triggered control for a class of discrete-time strict-feedback nonlinear systems whose model is accurately known.In order to solve the non-causal problem of the abovementioned system in the framework of backstepping,this thesis transforms the original system into a system in form of n step ahead predictor.Then,Lyapunov stability analysis is used to show the boundedness of closed-loop error systems and guarantee the tracking performance.Subsequently,this thesis further considers the design of model-based event-triggered control for discrete-time strict-feedback systems with unknown nonlinear dynamics.In order to solve the abovementioned causal contradiction problem,this part also converts the system to an n step ahead predictor form.Radial basis function neural network is used to deal with unknown nonlinear dynamics.In order to save more network resources,this thesis innovatively constructs an adaptive neural network model according to the transformed prediction system,and designs event-trigger control based on this transformed model.By using the Lyapunov stability technology,this thesis verifies the stability of the closed-loop system and the convergence of the tracking error can be verified.Finally,by combining the deterministic learning theory,the control performance is further optimized for strict-feedback nonlinear systems with unknown dynamics.Deterministic learning theory is used to store the knowledge of unknown dynamics,and the stored knowledge can be used to construct a controller when the same or similar task is performed in the future.As such,this thesis first learns the dynamic information of the considered system when network resources are sufficient.Then,when the network resources are limited,the stored constant neural network weights are used to construct a neural network model,and design event-trigger control scheme.Deterministic learning-based event-triggered control scheme has better transient performance,smaller steady-state error,and less network resource occupation than adaptive neural event-triggered control.For all control schemes aforementioned,simulation experiments are given in this thesis to verify their effectiveness.