Research On The Control System Using Multiple Models, Switchting And Tuning

The control of dynamical systems when there are large uncertainties is of greatinterest. Traditional adaptive control of such systems can not achieve good performace.Multiple models, switching and tuning approch is an efficient scheme to cope with suchproblems.In this thesis, we discussed the problem of adaptive control three of typessystems, determinstic system, stochastic system when random disturbances are present,and a nonlinear discrect-time system using neural networks.Firstly, the adptive control problem of a linear discrect-time system in nosie freecase is considered. The composition of the model set is discussed in detail.Based on theprediction errors of a finite nubmer of fixed and adaptive identification models,aprocedure is outlined for switching between a finite number of controllers to improveperformance.The thesis gives the proof of global stability of the overall system and thecovergence of the tracking error signal to zero in the determinstic case.Based on the previous foundations, the adptive control problem of a lineardiscrect-time system in stochastic case is considered.In this thesis we consider usingdifferent identification algorihms, and using the multiple models approach to select, online, the one that results in the best performace of the overall system for the givendisturbance characteristics.This thesis demonstrate that the convergence of theseschemes can be treated in a unified manner. Simulations are included to indicate theimprovement in the performace tha can be achieved using such schemes.Finally, adaptve control of a class of nonlinear discrect time dynamical systems iscosidered.The model set has two parts,one is linear adaptive model and another isnonliner model based on neural network.Each one has its own controller.At everyinstant of time,the performance of the linear identifier and nonliear identifer inpredicting the output of the plant are computed,based on a designed preformacecriterion the one which performs better is chosen to generate a certainty equivalencecontrol input to the plant.It is proved that improved performace and stability can be bothachieved.