A data driven subspace approach to predictive controller design

This paper shows the design of predictive controllers using the predictor, designed from the subspace matrices, obtained directly from the input/output data. The model-free design approach presented in the literature so far does not include all the important predictive control features such as inclusion of an integrator for offset-free control, constraint handling, feedforward option and a means of tuning the controllers through the disturbance model; these features are important for practical applications and hence, amongst other issues, addressed in this paper. The proposed predictive controller is demonstrated on multivariate systems using MATLAB simulations and an application on a pilot scale process. (C) 2002 Elsevier Science Ltd. All rights reserved.