Closed-loop Identification And Internal Model Control Methods Of Complex Multivariable System

The multivariable system in the industry has the characteristics of uncertainty,multi-time delay,strong coupling,and constraints in input and output,which makes it difficult to be accurately modeled.As a result,it is a challenge for the design of system controller to realize high-precision control of these systems using a traditional method.In order to improve the quality of system control,this thesis introduced advanced learning strategies,such as reinforcement learning,frequency domain analysis and average frequency domain nonsquare relative gain array(NRGA).With these strategies,several challenging problems in complex multivariable systems were investigated from the aspects of system model parameter estimation,internal model controller design and parameter optimization.The stability and robustness analysis of the internal model controller and the compensator design were also discussed.The derived solutions and improved strategies were validated by carrying out numerical simulations.The main research contents of this thesis include:1.A frequency response estimation(FRE)method was introduced to deal with the closed-loop system identification.Making use of the frequency characteristics of the system for the analysis of the control performance of the system,the parameter estimation for the identified controlled object model can be quickly and accurately accomplished.Through simulation analysis,it is yet found that this method has the limitation that the accuracy of the parameter estimation greatly depends on the selection of the attenuation factor.Facing this limitation,a frequency response estimation method based on the reinforcement learning framework(CARLA-FRE)was proposed by making use of the powerful online search and learning ability of the continuous action reinforcement learning automata(CARLA).The automata is able to give an optimized attenuation factor with the adaptive dynamic adjustment attenuation factor algorithm.After the tests of this method over a variety of basic function identifications,and the comparison between this method with the particle swarm optimization(PSO)algorithm and fireworks algorithm(FWA),it is revealed that the proposed algorithm has much stronger global search ability and accuracy.2.The extension of the proposed CARLA-FRE method to the closed-loop identification of multivariable square and non square systems to provide an optimization model for the design of subsequently advanced controllers.The extension employs the sequential excitation signal method to equivalently decompose a multiple-input multiple-output(MIMO)system into several single-input single-output(SISO)systems,and then uses CARLA-FRE to obtain the subsystem parameters.The derived analytical expressions are used to estimate the model parameters for the solution of the closed-loop identification problem of multivariable square systems and nonsquare systems.Finally,the method is applied to the internal model control of multivariable systems.The applications of the CARLA-FRE method in the internal model controller design for the multivariable square and nonsquare systems were verified by simulations with the classic Wood-Berry model and the Shell model,respectively.Incorporating reinforcement learning into the FRE method,this method has stronger online learning and anti-interference ability,and provides model support for the subsequent internal model control research.3.In order to improve the performance of multivariable and multi-delay control systems,the linear quadratic Gaussian(LQG)control method is used to design the optimal controller for the proposed model based on CARLA-FRE.Facing the noises,delays and partial parameter perturbations in the multivariable and multi-delay systems,this thesis introduced the LQG control method into the multivariate multi-delay process model to enhance the robustness and to restrain the turbulence of control by making effective compensation and uncertainty,parameter perturbations and system delays.4.The design of the internal model controller and the optimization of key parameters were carried out for a class of typical strongly coupled nonsquare systems.For the multivariable multi-delay rank-deficient system,the internal model controller based on penalty pseudo-inverse is designed.An internal model controller design method was proposed for the delay-rank system by introducing the penalty factor and using the pseudo-inverse of the non-full rank system to replace the inverse of the model.The continuous reinforcement learning identification method was used to find out the optimized and maximum penalty factor.For the multivariable rank-deficient system,the internal model controller design method based on the principle of compensator was proposed with the average frequency domain NRGA standards for the optimal selection of square subsystems.The simulation results show that the proposed method is not only simple and easy but also has strong robustness and stability in the case of system model mismatch.