Research On The Technology Of Data Driven Model Free Adaptive Control

With the development of science and technology,the industrial technology,equipment complexity and enterprise scale have been significantly advanced.The traditional control methods based on the precisely mathematical modeling,after been paid much attention to,have been found not quite suitable for industrial process controller design.At the same time,due to the development of hardware and information technology,for the enterprise now it is quite easy to acquire and deal with a lot of data coming from the industrial process.It becomes quite urgent to know how to realize the optimal control of complex production process directly using online and offline data when it's difficult to establish accurate mathematical models.On the basis of previous studies,this thesis is aimed at the model-free control of different types of nonlinear systems,including the scope from systems with low orders to ones with high orders,from continuous systems to discrete systems,and further research of data driven control technologies without precisely modelling.The main contents of this thesis are divided into five parts described as follows:(1)For a class of nonlinear systems with two orders,an extended state observer and a neural network observer are used to estimate the unknown model function and the external disturbance.Based on the unified form of observers from the both types and the design of a command filtered Backstepping controller,a new controller without knowning the mathematical model of control object is designed.By only using the input and output measurement information of the controlled object,the design of the model free controller for the nonlinear systems with two orders is realized.Finally,through the simulation of controlling a chaotic motion of a ship power system,the effect of the model free control method proposed is validated.(2)For a class of nonlinear systems with higher orders,a model-free control method based on a cascaded observer is proposed.In order to get the higher order differential information of the the controlled object output signal,a cascaded observer is designed and then a feedback controller using higher order differential signal is applied.The proposed control method does not need the accurate dynamic model,and the external disturbance is implied in the observed high order differential signal.The control method is also extended to a class of multi-input and multi-output nonlinear systems.Finally,simulation results show that the proposed model free control method is effective.(3)In consideration of the position and speed saturation in control input,the design of an observer based adaptive neural network constraint controller is presented.The adaptive neural network controller does not need to use the accurate mathematical control model,and a dynamic anti-windup algorithm is put forward to ensure that the output values of the controller do not reach the constraint threshold.The stability of the closed-loop system with the proposed method is analyzed Finally,based on both theory and simulation,it can be concluded that the proposed method is effective.(4)A new model free adaptive control algorithm is put forward using observer techniques to realize the PPD parameters estimation.The controller is found out to possess an internal model structure property,and is essentially an internal model controller.On this basis,the proposed algorithm is further improved using an improved feedback filter,which makes the system more robust against interference and ensures the system robustness is not decreased.For the whole closed loop system,its stability is analyzed using the Lyapunov stability theory.Finally,two simulation examples are given to verify the proposed algorithm,and the simulation results show that the proposed method is effective.(5)A RBF neural network is used to realize the on-line identification of the discrete nonlinear systems,and the identification results is used to get the Jacobian information of the controlled object,and then realize the online adjustment of the PID controller's parameters.An anti-saturation compensator is connected in series with the controller to ensure that the control signal is kept within the constraint range.The introduction of the anti-saturation module can still ensure that the parameter estimation errors are all bounded.Based on the proposed scheme,two adaptive PID controllers are proposed,which are the adaptive neural network PID control and the adaptive neural network PID-like control,and the appropriate stability analysis is given.The simulation results show that the method is effective and is better than the conventional PID control.In this paper also propose a PID control algorithm based on the improved learning theory.After obtaining the estimated model by using the improved real-time learning method,the corresponding Jacobian information of the original system can be replaced by the Jacobian information of the estimated model,and an adaptive updating law of the PID parameters can be obtained by using an optimization algorithm.Finally,the effectiveness of the proposed method is verified by simulations,and the results show that the proposed method is effective in the situation of complete data and partly effective in the situation of incomplete data.