Research On Multiple Model Adaptive Control For Nonlinear Systems

In actual industrial production,complex nonlinear systems often have one or more nonlinear components,and often accompanied by external interference,subsystem dynamic changes,system parameters jump and other problems,which will make it difficult to control the system.Traditional adaptive control methods can not effectively control such complex nonlinear systems,and the multiple model adaptive control(MMAC)method provides a good solution for the control of such complex nonlinear systems.Therefore,this thesis uses the theory of MMAC method to solve the control problems of unknown system parameters and system parameters jump for a class of complex nonlinear systems.And the problem that too many sub-models in MMAC lead to the increase of system computation.The specific research is as follows:1.For a class of nonlinear discrete-time dynamic systems with unknown parameters,a new MMAC method based on neural network is proposed.Firstly,the system is divided into linear and nonlinear parts.For the linear part of the system,several fixed models based on localization method can cover the parameter range of the system.On this basis,an adaptive model is established to improve the performance of the system.For the nonlinear part of the system,a nonlinear neural network prediction model is established to approximate the nonlinearity of the system.Then,the corresponding controller is designed for each sub-model.Finally,a performance index function based on error norm is designed to switch the controller.The simulation results show that the proposed MMAC method can significantly improve the transient performance of the nonlinear system compared with the traditional MMAC method with uniform distribution in the parameter space.2.For a class of nonlinear discrete-time dynamic systems with parameter jump,a MMAC method based on clustering method and neural network is proposed.Firstly,the fuzzy c-means clustering algorithm is used to classify the prior data of the system,then we use RLS algorithm to build multiple fixed models for each type of data.On this basis,two adaptive models are established to improve the response speed and control quality of the system,and a neural network prediction model is established to compensate for the nonlinearity of the system.Finally,the performance switching index based on signal boundedness and measurement error is used to switch the controller,and the stability of the closed-loop system is proved.The simulation results show that the proposed algorithm can better solve theparameter jump problem of the nonlinear system and make the system have good control quality.3.For the optimization of model base in MMAC method,Considering the actual operation data of the system,a dynamic optimization model base method based on similarity criterion and maximum number of models is proposed.This method can comprehensively consider the new data and determine whether the data should be incorporated into the sub-model modeling,and by setting the maximum number of models to ensure that the system with the minimum number of sub-models can ensure the control performance of the system.The simulation results show that the proposed algorithm can greatly reduce the number of sub-models and has good control effect.