Técnicas de identificação voltadas para a otimização de processos em tempo real

Abstract

This thesis reviews the subject of static real-time optimization and presents the elements of various methodologies and industrial implementations reported in the literature. The central role of process identification, model accuracy and availability of measurements in real-time optimization is shown. Special attention is given to the basic aspects of the practical reality of processes operation, such as the identification of processes, given the available measured variables, and the appropriate treatment of the involved numerical aspects. Considering industrial processes with massive availability of data, a strategy was proposed to monitor large-scale processes based on the use of empirical models combined with methods of statistical process control, indicating the quality of the model and being effective for fault detection and diagnosis. Processes with low availability of information are also considered with the development of an on-line identification procedure, which aims to deal with the difficulty of building and updating models for use in real-time procedures when several important quantities are not available. The performance assessment of different techniques used to solve the optimization problem revealed possible vulnerabilities to solve data reconciliation problems, even when rigorous models and adequate measured variables are available. The studies contribute to the subject of real-time optimization as they represent components of commercial RTO systems. They also represent innovative decision support techniques with broader application, as they can be implemented with lower investments and do not demand the complexity required by RTO systems.