Study Of Model Predictive Control Methods Based On Subspace Identification

It is urgent to find appropriate identification and control methods for the modern industrial processes. Some important problems remain unsolved for the complicated multivariable systems. Model predictive control has a lot of advantages in dealing with multivariable processes. Practical applications show that it can achieve good performance. Traditional industrial predictive control uses input/output model, including both parameter and non-parametric ones. However, in order to obtain further improvement of control performance, both academia and industry think that it should adopt the state space model. For this purpose, the modern filtering theory and controller design methods, which are developed in recent years, can play a role. Subspace identification method uses the process input and output data to obtain the state space model directly. The intermediate result of identification process, which is always called subspace predictor, can be directly used to predict the output. The main works and contributions of this thesis are as follows:â‘ A subspace identification algorithm, which obtains the state space model from the input-output data containing noise sequence, is introduced. This method also provides a solution when the noise is non-stationary. The efficiency of this method is illustrated by two simulation examples, a CD player arm and a CSTR.â‘¡A state space predictive control method, based on subspace identification, is proposed. The state space model, obtained through the subspace identification algorithm, is regarded as the system model. Unconstrained and constrained predictive control algorithms are listed, respectively. By taking a CD player arm system as an example, the goal is to achieve the system output tracking control. The simulation results show that the control effect of this method is satisfactory.â‘¢In order to deal with nonlinear and time-varying characteristics in the practical industrial processes, an adaptive predictive control method, based on subspace identification, is proposed. Through on-line updating the R matrix with receding window, the new prediction model parameter matrices are obtained. By comparing the prediction error before and after updating, it considers whether or not to update the prediction model. This control method is applied to the process control simulation on a 2-CSTR. Through comparisons of its performance with PID controller and a fuzzy control scheme, the superiority of the proposed control method is illustrated.