FAST ADAPTIVE GENERALIZED PREDICTIVE CONTROL FOR SYSTEMS WITH VARIABLE PARAMETERS

Abstract

This paper proposes a fast predictive control structure with online model update according to process parametric variations. The proposed controller is based on the Generalized Predictive Control (GPC) algorithm, but it integrates the recursive least squares identification method with a variable forgetting factor to estimate at each iteration the parameters of a linear structure model used for multi-step ahead prediction. For a system with constraints on the process variables, the resulting optimization problem of GPC is solved using quadratic programming based on the Alternate Direction Method of Multipliers, which allows the control signal to be obtained with small computational effort. In order to validate the proposed algorithm an experimental case study that considers the speed control of a direct current motor and the proposed controller embedded in a microcontroller STM32F303K8T6 is presented. Experimental results use as baseline the GPC with fixed model parameters and show that the proposed fast adaptive predictive control structure is able to keep almost the same transient response for all the considered operating points of the motor, while GPC presents high oscillations at operating conditions far from the one used to obtain the nominal model. Even though the proposed controller needs to solve two optimization problems at each sampling instant, it can run about 60 times in a second in the microcontroller used in this study