Real-time optimiztion with persistent parameter adaptation using online parameter estimation.

• Ano de defesa: 2018
• Autor(a) principal: Matias, José Otávio Assumpção
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Detecção de estado estacionário; Estimação em tempo real; Online estimation; Otimização global da planta; Plant-wide optimization; Real-time optimization; Steady-state detection; Tempo-real (Otimização)
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/3/3137/tde-17122018-084058/

In standard Real-time Optimization (RTO) implementations, the plant needs to be suciently steady in order to update the RTO model parameters reliably. However, this condition is seldom found in practice. Moreover, because the RTO model is only updated when the plant enters a stationary condition, the optimizer is likely to be out of phase with highly perturbed plants. The main contribution of the thesis is the proposal of an alternative RTO approach, called Real-time Optimization with Persistent Adaptation (ROPA), which integrates on-line parameter estimation in the optimization cycle, avoiding the steady-state (SS) detection step. Instead of predicting the SS, the online estimator keeps the model up-to-date with the plant and allows running the economic optimization at any time, even instants after implementing the current RTO decisions. ROPA provides an intermediary solution between static and dynamic optimization schemes. While it approximates the optimal trajectory, ROPA enables the use of well-established static RTO commercial solutions. Furthermore, the new approach is the key for decoupling the model estimation problem in order to achieve plant-wide optimization. Another contribution of the thesis is to provide several case studies in which ROPA is tested and compared with the standard RTO implementation: a Williams-Otto reactor, a Fluid Catalyst Cracking unit and a separation-reaction system. The idea is to illustrate ROPA convergence properties and how the plant-wide optimum is achieved by asynchronously updating the global plant model. The results show that ROPA is able to track the stationary (plant-wide) optimum. In addition, they conrm that the renement of the prediction capacity, by decreasing the time between two sequential optimization, enhances the disturbance detection of the optimization cycle and leads to a better economic performance.