Research On Adaptive Model Predictive Control Based On Subspace Identification

Nowadays most of the control methods are based on mathematical model. But the actual process more complex, it is nearly impossible to build the mathematical model from mechanism. The demands of modeling process by system identification are focused. The identification based subspace method has been paid more attention especially for multi-variables process in the last two decades. In order to realize the online identification of state-space models,a new recursive subspace identification method is proposed, integrating the method with Predictive Control, this paper mainly contains several aspects followed:(1) By the subspace identification and model predictive control algorithm combining, proposed model predictive control algorithm based on subspace identification. Subspace matrixwL anduL obtained in the middle of the identification process. Binding model predictive control selected objective function, the subspace predictor with quadratic model objective function combined, and constructed objective function of predictive control on subspace identification. The objective function is minimized to obtain control law. In order to eliminate the steady-state errors, with the integral in the controller. Through comparisons of its performance with PID controller and DMC control scheme, the superiority of the proposed control method is illustrated.(2)In order to deal with nonlinear and time-varying characteristics in the practical industrial processes, an adaptive predictive control method based on recursive subspace identification, is proposed. Through on-line updating the R matrix with recursive, the new prediction model parameter matrices are obtained. The efficiency of this method is illustrated by simulation example, the quadruple-tank.(3) With the passage of time, the calculation amount will continue to increase. In order to reduce the amount of computation, the paper proposes a selective adaptive subspace control algorithm which is based on lazy learning. To reduce the computational burden, we can use a criterion which can add data selectively and reduce the unnecessary calculated amount greatly...