Data-driven Disturbance Observer And Self-learning Control For Nonlinear Systems

In this paper,a series of iterative learning observers and observer-based data-driven iterative learning control methods are proposed to solve the problems of external disturbances,random measurement noise,and nonrepetitive initial values.The stability of the proposed method is discussed and analyzed,and strict mathematical proof and simulation verification are given.The main innovations of the paper are summarized as follows:Firstly,two data-driven iteration learning based accumulative disturbance observer(ILADOB)are designed for linear systems with nonrepetitive external disturbances,in order to achieve the purpose of disturbance estimation under the data-driven design and analysis framework.If the system state is measurable,the proposed state-based ILADOB is applied,otherwise,we propose an output-based ILADOB.The two proposed methods are estimated along the iteration axis in a finite time interval,and the disturbance can be estimated by using the system data in the previous iteration.The results are further extended to nonlinear nonaffine systems.Theoretical analysis and simulation results show the effectiveness and applicability of the two methods.Secondly,for repeatable nonlinear nonaffine systems with nonrepetitive external disturbance,a disturbance observer-based data-driven iterative learning control method is proposed.By introducing the iterative learning disturbance observer,the total disturbance of the system caused by external disturbance is estimated and used as a compensation in the iterative learning controller.Therefore,the proposed method can effectively reduce the influence of nonrepetitive disturbances on the system and improve the control effect.Both theoretical analysis and simulation results verify the effectiveness and applicability of the proposed method.Thirdly,for nonlinear nonaffine system with nonrepetitive initial values and external disturbances,a Luenberger observer-based iterative learning control scheme is proposed under the data-driven framework to improve the robustness of the controlled system.The observer is used to estimate the system output,which is used as compensation in the controller directly,so as to reduce the influence of nonrepetitive uncertainties.The proposed observer is updated along the iterative axis,and the data in previous batches can be used to enhance the performance of the observer.Strict theoretical analysis and MATLAB simulation research prove the effectiveness of the proposed control algorithm.Fourthly,an observer-based iterative learning control method for a nonlinear nonaffine multi-agent system(MAS)is proposed to achieve consistent control of MAS under the influence of nonrepetitive initial values and measurement noise.Firstly,the input-output(I/O)relationship of the nonlinear nonaffine MAS system along the iterative axis is described as a linear data model,and the effects of nonrepetitive initial values and measurement noise on system dynamics are transformed as a total disturbance in a linear data model,and then an iterative disturbance observer is designed to estimate it.Further,an observer-based switched iterative learning control method(OBSILC)is proposed,which includes two iterative learning control protocols and a parameter adaptive estimation law,which both use disturbance estimation as compensation.The difference between the two iterative learning protocols is that one of them introduces an iterative decay operator to further reduce the influence of uncertainties on control performance.The two iterative learning consistency algorithms switch according to a preset error threshold.Stability analysis and simulation studies prove the effectiveness of the proposed OBSILC.