Control-Oriented Iterative Identification And Control Design

With the development of control theory, many of advanced control methods, such as robust control, adaptive control, model prediction control etc., are widely applied in practice. These methods usually depend on the accurate plant models which are mostly obtained via system identification methods. The classical identification methods always seek for a deterministic model to approximate the real plant model. However, the classical identification and controller design, which both depend on their own independent optimal criterion respectively, are consistent superficially but separate essentially, and the obtained controller on the basis of the identified model will not always ensure the stability and the desired control performance. The control-oriented identification is a modeling process with the aim to get the desirable controller, and the quality of identified model is evaluated by control performance. Therefore, this thesis will focus on the control-oriented iterative identification and control design method with the purpose of integrating identification with control design to achieve fast and accurate model identification and controller design.This thesis contains the following main sections:First of all, excitation signal optimal design for closed-loop system identification is investigated for selecting optimal excitation signal. The problem of excitation signal selection is transferred into an optimization problem to find the optimal signal spectrum. Its objective function is defined as the joint approximation of input and output. The constraints of this optimal problem include two parts: the one is the robust stability of closed-loop identification, which will ensure that the initial control can stabilize the identified model, and the other is the power restrictions of excitation signal, control signal and output signal, and these power restrictions are transfered into LMI that can be solved by optimization methods. On the basis of these, optimal excitation signal design problem is formulated. Finally, the adaptive iterative method is proposed and its convergence is analyzed to deal with the complicated problem that the excitation signal typically depends on unknown system model in optimal excitation design. Simulation result shows that identified model will be improved by using the excitation signal optimal methodSecond, the iterative identification and control design method is investigated to combine the identification with control design. The uncertainty model of the real plant is expressed by prediction error (PE) uncertainty model structure which is obtained by system identification, and the uncertainty model set is described by the Vinncombe distance. According to these, the closed-loop stability condition of the uncertainty system in the worst case obtained by the frequency dependent stability margin is proposed to ensure the closed-loop system stability during the iterative procedure. And, the condition of closed-loop system performance improvement in iterative procedure is proposed to guarantee a better close-loop system performance. According to the above two conditions, the iterative controller design method and the whole procedure for iterative identification and control design are developed. This method will improve the stability and convergence of classical iterative methods, and ensure the stability of closed-loop system and control performance in iterative procedure. Finally, a Matlab/Simulink platform is established to verify the proposed method. Simulation results show the good performance of the proposed method.Finally, model validation methods are investiaged to test the quality of identified model. The whole model validation procedure is divided into two different parts. The first part is the model structure validation. For that, a model validation method based on residuals cross-correlation test is presented to validate identified model with unmodelled dynamics. In open-loop condition, the model validation is firstly transferred into a hypothesis validation problem, and a new residual estimation method is proposed. Through analyzing the cross-correlation function of residual signal and input signal, model validation is implemented. The second part is a closed-loop model validation based on the model error modeling. In the case of the controller with non-minimum phase zeros, reasonable model validation method is analyzed, validation procedure is presented, and model validation result with probability is given. Simulation result shows the proposed method is effective.