Research On The Control Method For Uncertainty Parameter System

Uncertainty is widespread in all types of systems,and automatic control systems are no exception.These uncertainties come from external disturbances,internal connections and coupling,subsystem faults,parameter fluctuations and so on.The existence of uncertainties seriously affects the normal operation of the system.Since the emergence of control theory,the study of uncertainty has never stopped.Feedback control,adaptive control,robust control and system identification are typical ways to solve different kinds of uncertainties.Attempts to achieve a certain balance between the indicators of the system utilizing control and identification,known as dual adaptive control,was listed as one of the 25 problems that had a great influence on control theory in the last century by the IEEE Control Systems Magazine in 2000.However,this open problem with great theoretical significance and practical value has not been satisfactorily solved so far.At present,two learning strategies have been developed to solve this problem:active learning strategy and passive learning strategy.The existing works propose some sub-optimal control methods but avoid the mutual influence between system variables.Their common disadvantage is that they artificially deprive the right of active detection to control the future,so that the obtained sub-optimal control has only the characteristics of passive learning.Aiming at the dual adaptive control problem of uncertain systems,a new method of controller design is proposed in order to eliminate or reduce the uncertainties in the system,so that the system can run in an optimal way and achieve the expected performance.The main research results of this paper are as follows:For the control problem of uncertain systems caused by unknown parameters in the model,the traditional method is divided into two stages:the first stage is to collect input-output data of the system and identify the system model;the second stage is to design the so-called optimal control by using the identified model and index requirements.However,dual adaptive control integrates the two stages and adopts the strategy of one-side control and one-side identification.The simulation results show that the dual adaptive control is superior to the traditional control,and then a general framework of the dual adaptive control design is proposed.Since dual adaptive control is superior to traditional control strategy,what are its characteristics and internal regularity?Referring to the exploratory-utilizing framework of reinforcement learning,this paper obtains the exploratory-cautious effect of dual adaptive control.With LQG as the carrier,a dual adaptive control is designed to solve the uncertainty problems caused by unknown parameters,environmental disturbances,measurement noise and initial state in the model.This control not only enables the control system to have the desired secondary performance index,but also learns the minimum interval containing unknown parameters with the given precision in advance.In order to find out the internal regularity of large data,aiming at the autoregressive sliding model with unknown parameters and aiming at one-step optimization,the learning index closely related to learning is derived.Combining the control objective with the learning objective,the controller design method with learning characteristics is obtained.In view of the short-sighted behavior of one-step control target,under the constraint that the state variables can be accurately measured,this paper obtains the adaptive dual control,which makes the whole system optimal.On the one hand,this control can make the system run in the optimal direction;on the other hand,it can estimate the unknown parameters and reduce the uncertainty of the system.