Controle preditivo de sistemas híbridos

Abstract

The industry's need to improve aspects of production such as quality and eciency meant that techniques and devices for more ecient control were adopted in view of the dierences that occur in systems. most systems are not only characterized by the continuous dynamics usually applied to describe this behavior but an association with this dynamic elements with discrete characteristics (logical). For this systems relates to the expression hybrid systems whose dynamics processes characterize the behavior continuous real-time system associated with a discrete event. The control of hybrid systems requires answers in accordance with the presence of discrete event and continuous interacting that makes control strategies increasingly sophisticated are developed. In this work, was adopted the hybrid MLD formalism (Mixed Logical Dynamical ) for the representation of hybrid systems using dynamic logical propositions that express such a system in the form of linear constraints of binary variables interacting with the continuous behavior of dierential equations. In this context, the predictive controller based on an optimal model developed from the selected formalism is suggested as an option for the control of hybrid systems. In this work, the qualitative modeling of discrete systems using the MLD formalism is presented for which it is developed to check these systems to evaluate the hybrid modeling. These techniques were applied to simple problems in chemical engineering as systems of three tanks connected in series and a CSTR reactor with heating. In each of the systems were made simulations coupled to qualitative models in various scenarios and compared to the continuous model of dierential equations and their discretized version. The action of the MPC controller for hybrid systems was studied also, developing it from the MLD model and presenting several simulation scenarios to investigate the eects of control over the system. Such control is developed to the l1 and l2 norms, enabling a comparison between the two control options and the comparison between these controllers with classics like PI. The results for the model in the MLD formalism are satisfactory and consistent with the real behavior of the system. The control problems developed show intervention MPC controller more ecient than the classical PI controller associated with the control on/o mainly by the choice of variables to be manipulated. The comparison between controllers l1-MPC and l2-MPC gives results that indicate similar control actions for each of these controllers depending on the location of the optimal point found in solving the optimization model for the controlled system.