Identificação relevante ao controle preditivo - método EN-PH de otimização numérica baseado em técnicas de regularização.

Abstract

Unlike traditional controllers, predictive controllers implement control actions to be applied to the system from predictions of process outputs at a certain future time interval. This is done so that the controller anticipates future output variations so that it can act more efficiently. The most straightforward approach to obtaining appropriate models for Predictive control is to define and minimize the cost function of multiple steps ahead. This can be achieved using a non-convex optimization method such as Levenberg-Marquardt's. In the PLS-PH method, the nonlinear optimization problem is solved from a linear search and the transformation of predictors to the latent variable space (PLA, 2012). In this work a new multivariable MRI identification method called EN-PH (Elasti -Net Prediction Horizon) was proposed. It consists of a numerical optimization algorithm associated with the Elasti-Net regularized regression technique. The new method proved to be able to estimate the process model parameters efficiently, even in the presence of collinearity between the predictors.