Controle preditivo distribuído para sistemas não lineares com particionamento automático

Abstract

One of the main purposes of a control system of an industrial plant is to guarantee closedloop stability and to coordinate the various existing interactions between its subsystems. Subsystems of a plant are usually designed independently or added later with the evolution of the installed plant. These changes usually occur motivated by production requirements or environmental regulations. Most large-scale systems implement the decentralized control as the control strategy of choice. However, for subsystems with strong interactions, this approach can lead to unacceptable performance. Furthermore, centralized control is able to address optimally the problem of interaction, but with high structural and organizational costs, making costly such a complex structure and upgrade maintenance. A structure that preserves the topology and flexibility of decentralized control and at the same time may offer a nominal closed-loop stability guarantee is the distributed control approach. In this control structure, the interactions between subsystems are modeled and information between the subsystems is shared between them. The present work presents contributions in the study of distributed controllers for nonlinear and large scale systems. Four types of Distributed Model Predictive Control (DMPC) are proposed: non-cooperative locally linearized DMPC, cooperative locally linearized DMPC, non-cooperative non-linear DMPC and cooperative non-linear DMPC. The four proposed controllers are for application in non-linear processes, but the first two use the locally linearized model to predict the outputs and the last two make direct use of the nonlinear model. The proposed controllers were evaluated in three different case studies and their performance compared to centralized and decentralized control strategies and satisfactory performance was obtained for the developed controllers.