Controle preditivo não linear com estratégia adaptativa baseado em simulated annealing acoplado paralelo

Abstract

Model Predictive Control (MPC) is an advanced control technique that has had a significant impact on process industries. There are several reasons for its great acceptance, e.g., it copes with multivariable control problems naturally, it considers constraints on input and output variables, and adapts to structural changes. Despite the growing research effort focused on the development of nonlinear predictive control strategies, the use of these techniques in real systems is still a challenge, because the developed algorithms are usually more complex and, sometimes, do not allow real-time applications for fast dynamics systems. Nevertheless, the constant increase in the speed and power of computing makes this perspective real. This way, the purpose of this work is to develop a nonlinear predictive control strategy that uses a stochastic optimization algorithm with great parallel scalability known as Coupled Simulated Annealing (CSA). The strategy aims at solving directly, without approximations of the process model, and in a parallel way, the optimization problem associated to nonlinear MPC for real-time applications in fast dynamics systems. As the model used in predictive control is only a mathematical approximation of the plant, there is a possibility of mismatch between their behaviours. Thus, the proposed strategy seeks to meet the requirement of robustness to model uncertainties, solving the problem adaptively. In the present work, the parallel version of CSA was used to solve the constrained control problem in three distinct nonlinear systems: real application in a Coupled Tanks system, simulation of rotary inverted pendulum control and a nonlinear chemical reactor, considering the mismatch between plant and model parameters. The experiments results showed the efficiency and characteristics of the control strategy to control these faster dynamics systems, requiring few adjustments to be applied in the different control problems, besides the advantages of optimizer parameters initialization. The efficiency and characteristics of the adaptive strategy considering the mismatch between model and process are presented to simulate the control of a nonlinear chemical reactor with uncertain parameters.