Planejamento de sinais para identificação de modelos multivariáveis com restrição

Abstract

A new methodology for the identification of empirical models for Model Predictive Control (MPC), which considers both operational and phenomenological constraints, is proposed in this dissertation. From the work of ORENSTEIN (2013), a "gray box" method of identification type was developed capable to generate only physically consistent models, i.e., models in which the value and the signal of the static gains have physical sense. The developed methodology, which makes use of step-type as well as GBN-type (Generalized Binary Noise) input disturbances, involves both online and offline steps, and the offline steps are performed by a computational package composed of four algorithms: the first two ones perform the data analysis of the process and the last two ones solve optimization problems with restriction. The methodology was applied in the identification of linear dynamic systems and of a classic problem in this field, namely, a Shell distillation column, showing to be fast and robust in the simulations presented. The simulation times were used as indicators of speed and the statistical parameters MRSE (Mean Relative Squared Error) and MVAF (Mean Variance-Accounted-For) as indicators of robustness of the proposed methodology.